An adaptive RFID anti-collision algorithm for network intrusion detection

Abstract

Radio frequency identification (RFID) provides real-time network monitoring capabilities for threat identification. However, accurate detection is impeded by tag interference. This paper presents an adaptive collision tree algorithm that selects optimal binary or octal splits based on collision counts to handle interference. Experiments demonstrate an integrated RFID intrusion detection framework that achieves 8.98% higher throughput and 99.82% detection accuracy compared to other protocols. The method enables efficient real-time threat identification as networks proliferate. However, there are limitations to the approach, such as assumptions of fixed tag populations rather than dynamic tags and a lack of field testing. To strengthen the approach, further research on fluctuating tags and validation in real-world network deployments is necessary. This work presents an adaptive method for leveraging RFID to achieve scalable and accurate network intrusion detection.

Keywords

Collision tree radio frequency identification electronic tag network intrusion identification

1 Introduction

The Internet of Things (IoT) fosters Internet connectivity, which can be achieved through a range of information sensing devices, enabling specific targeting and identification. Radio Frequency Identification (RFID) technology is a frequently used method of tag identification. When a large amount of data needs to be managed, the intelligent control capabilities of the system can be utilized. Therefore, RFID technology has been extensively researched and implemented, significantly easing the daily life and work of individuals. It has numerous applications, including traffic and logistics management, safety and security, and various retail markets (Baygin et al., 2022; Pan et al., 2023; Saghlatoon et al., 2020). In addition to its numerous performance benefits, this technology boasts an automated and non-contact realization, requiring no human involvement. It can swiftly identify targets across multi-targeted scenarios, is applicable to a wide range of fields, has minimal environmental restrictions, and possesses a large capacity for tag storage. Furthermore, its storage capacity will continue to improve in subsequent advancements. The superior functionality and outstanding performance of RFID technology have established its position in the application market (Zhao et al., 2022). In the current era of information abundance, every target tag within the network is assigned an individual serial number for precise identification. However, this also results in heightened complexity for network processing. Therefore, a technology that can identify multiple targets is necessary, and RFID technology aligns with this requirement. However, the primary area for improvement in RFID technology is the susceptibility to data interference issues when multiple tags respond to the same reader, which remains a crucial shortcoming that requires further optimization and resolution (Srivastava et al., 2020). RFID technology is commonly used in supply chain management, inventory control, and other fields due to its efficient tracking and monitoring capabilities. In the network security environment, RFID can monitor and identify abnormal behavior in the network, improving the system’s ability to identify potential threats. However, in RFID systems, each tag responds to the reader’s query via a wireless signal. When multiple tags are within range of the reader simultaneously, they may respond to the reader’s signal simultaneously. As a result, signals from multiple tags may collide in the air, making it impossible for readers to distinguish or correctly interpret information from any single tag. This phenomenon is comparable to ‘packet collision’ in wireless communications, which can result in data transmission errors or failures. Additionally, environmental factors can also affect the signal propagation of the system. For instance, materials such as metal objects and liquids can absorb or reflect wireless signals, thereby interfering with the signal reception of the reader. The anti-collision algorithm can effectively manage the signals of multiple RFID tags, reducing data reading errors and signal conflicts. By combining RFID technology and efficient anti-collision algorithms, it can improve the accuracy and efficiency of the model, reduce false positives and missing reports, and make the network environment more secure and stable. The integration of Network Intrusion Detection (NID) and RFID technology is deemed necessary and aligned with current demands. However, the ongoing research in this field is not yet sufficiently developed, and optimization of the method used for tag collision in RFID technology is necessary. The study presents an NID method that utilizes the Adaptive Multidimensional Collision Tree (AMCT) RFID algorithm. In order to facilitate the rapid reading and identification of multiple RFID tags in the IoT, the traditional anti-collision algorithm, which lacked sufficient throughput and efficiency, has been replaced. A multi-digit fork approach enables more accurate prediction of label responses and reduces unnecessary query steps. The algorithm dynamically adjusts the fork mode based on real-time collision information, thereby increasing flexibility in dealing with various collision scenarios. Overall, the research algorithm enhances the processing efficiency of RFID technology in mass label recognition, demonstrating broad potential for application. This serves as a reference for integrating RFID technology and the IoT in the future. The method is divided into four sections. Firstly, the current research status of NID and RFID anti-collision algorithms are introduced. Secondly, the NID recognition technique developed through the AMCT RFID algorithm is designed. Thirdly, the reliability of the proposed recognition algorithm is verified with experiments. Finally, the experimental data is summarized.

NID is currently a popular research trend. Kan X et al. introduced an innovative adaptive convolutional neural network optimized by a particle swarm algorithm utilizing differing inertia weights to automatically optimize the structural parameters of a one-dimensional network, and implementing a fitness value in the form of a cross-entropy loss function. By comparison to other algorithms, their method displayed effective and reliable performance in a range of experiments (Kan et al., 2021). Martin ML and colleagues addressed the issue of high-dimensional network address features in machine learning models. They proposed a new feature that utilizes the distance between source and destination addresses and obtained an embedded representation of the address using a neural network. The authors compressed the address values by combining them with a hash function. Training the model in a self-supervised learning framework and applying the new features to the classifier significantly improved the accuracy of NIDs (Martin et al., 2021). Ghasemi F et al. proposed a lightweight RFID based authentication technique using stream encryption to protect privacy between legitimate components and various keystream generators to construct a nonlinear pseudo-random number generator that provides forward security and protection against multiple attacks. The impact of different error control techniques in the communication BER was examined and the advantages of the method in terms of cost, storage space and communication cost were demonstrated (Ghasemi & Babaie, 2022). RFID technology through tag identification has higher novelty and comprehensive capability in NID research compared to the traditional algorithmic network model. According to Chen X et al., RFID technology was advancing quickly. They proposed a highly dynamic RFID lost tag identification system based on the HDMI protocol, which simultaneously identifies lost tags by combining the reply bit and the time slot position of the tag replies. The system’s efficacy was experimentally verified (Chen et al., 2022).

However, RFID technology has a significant defect that when multiple tags respond to the same reader, based on the phenomenon of tag collision, resulting in the final recognition failure. Ai Y et al. proposed anti-collision algorithms capable of solving the tag interference problem of RFID, based on the time-slot ALOHA method with the regression binary search tree for the tags within the collision time-slots, and in the case of the same number of tags, their algorithm requires less time slots, high recognition efficiency and short time (Ai et al., 2022). Equal area division was used by Huangshui HU et al. to implement an RFID anti-collision strategy. This meant that all tags in the read-write receiving range were grouped according to the equal area form, and that the number of tags in each group was simultaneously estimated using dynamic prediction weights. The identification process was then carried out in accordance with the ideal number of time slots to modify the scheme. The experimental results demonstrated that their method effectively increases the system’s throughput rate (Hu et al., 2020). To get the unread tag information, Wang C et al. suggested a binary search tree approach based on physical layer network coding (Wang et al., 2021). This algorithm involved pressing conflicting signal information into the stack and decoding the physical layer network coding. Radi J and colleagues proposed a collision avoidance algorithm that utilizes the dynamic frame timeslot ALOHA. The algorithm aims to enhance reading efficiency while accurately estimating the remaining number of tags. To achieve this, the authors introduced an estimator that can be generalized to arbitrary frame lengths. This reduced complexity in the estimation problem and relies solely on the number of conflicts, rather than the number of conflicting and successful timeslots (Radi, Oli & Kiljo, 2020).

RFID technology has become essential for the development of IoT, but little research has been done on combining it with NID. Furthermore, multi-tag interference presents an issue. As a solution, an AMCT-based RFID anti-collision algorithm has been proposed and applied to NID resulting in a more accurate and efficient network security maintenance.

2 Research method

To solve the problem of network security aggravated by data explosion, the study proposes to use RFID technology for tag identification to solve the Network Intrusion (NI) problem. Firstly, the basic concept and composition structure of RFID technology are introduced, and the more appropriate collision tree (CT) algorithm is selected to deal with the data interference phenomenon. Then the CT-based AMCT optimization algorithm is proposed and designed for the growing tag volume situation.

2.1 RFID technology model for NI tag recognition

The network era of rapid data growth has put forward higher requirements for network security maintenance.NID is designed to identify external attacks as well as internal misuse of resources, which can be identified by tags, such as a lack of tags or an imbalance, etc (Chen et al., 2021; Zhang et al., 2020). The fundamental structure of RFID, a popular data transmission technology with automation, high speed, and high accuracy benefits, is depicted in Fig. 1. RFID allows tags and readers to communicate without physical contact.

Fig. 1

Basic structure of RFID system.

Among them, the read-write can realize the communication between the electronic label and the data, belongs to the information processing centre. The electronic label, also known as the radio frequency label, is responsible for recognition, response, and read-write functions for various commands. It carries a vast amount of data and has a performance index that includes read-write speed, storage size, and anti-collision ability. The data processing system is the data repository in the background, which can be scheduled, integrated, etc., and the middleware is the communication bridge between this system and the reader (Abbasian & Safkhani, 2020). The problem of rising number of tags brought about by the development of the network has led to the emergence of many pending defects in RFID as well, such as when multiple tags are propagated in the same reader, data interference often occurs, resulting in tags that cannot be recognized, which requires RFID to have a strong anti-collision performance. It also needs to have the ability to efficiently identify multiple reader signals, as well as the ability to identify foreign tags. According to Fig. 2, the common anti-collision techniques include time division multiple access, space division multiple access, code division multiple access, and frequency division multiple access.

Fig. 2

Common anti-collision strategies.

Division of frequencies Due to its high cost, multiple access is not as widely used as it could be. To achieve collision avoidance, it uses frequency classification by separating sub-channels. The code division multiple access concept is categorized based on the data’s coding sequence. While this has a better effect, it is still excessively complex. The spatial division of the space division of multiple access likewise results in the same high costs and design complexity issues (Samsami & Yasrebi, 2021). Time division multiple access is divided into time blocks, making the structure relatively simple and cost-effective, hence its widespread use in RFID collision avoidance technology. Classic time division multiple access algorithms include the ALOHA protocol algorithm and the tree-based algorithm. The former belongs to the uncertainty algorithm, and tag starvation problem often occurs, so the study selects the latter CT algorithm to achieve anti-collision, as shown in Fig. 3.

Fig. 3

Operation mechanism of CT anti-collision algorithm.

First of all, it needs to read-write to the tag to send commands and response. When the number of the corresponding tag is 1, there is no collision phenomenon, it can directly complete the communication. However, to create a new query prefix that will be embedded in the stack, it is required to determine the bit position where the collision occurs and to maintain the collision for all previous high positions following a multi-tag response collision. Subsequently, all the tag serial numbers are compared with this prefix, and the tags with the same prefix start responding to identify whether there is a collision or not, and the above steps are repeated until the number of query stacks is zero (Zhou & Jiang, 2021). It can be seen that the CT system contains only time slots for collision and smooth recognition, and node splitting exists only at collision. Therefore, its tree structure is a full binary tree, and the searches T_CT required to recognize a tag is shown in Equation (1). $T_{CT} (N) = 2 N - 1$ (1)

In Equation (1), N is the total number of tags. The throughput rate S_CT of CT algorithm is shown in Equation (2). $S_{CT} = \frac{N}{2 N - 1} * 100 % \approx 50 %$ (2)

A throughput rate that is indefinitely close to 50% is achieved through an increase in the number of tags. The tag’s ID serial number is a binary sequence, with each bit possessing specificity and mutual exclusivity. This introduces the concept of binary determinism. By utilizing this feature, the reader can perform the query directly without generating a new prefix. The number of algorithmic queries under this step is shown in Equation (3) (Li et al., 2022). $\underset{CT}{T^{'}} (N) = 2 N - 1 - 2 M$ (3)

In Equation (3), M is the number of collisions that occurred and $\underset{CT}{T^{'}} (N)$ is the number of algorithmic queries after the introduction of the binary deterministic concept.

2.2 Application of AMCT-CT in RFID-NID

Since the CT algorithm splits only two branches to cope with signal collision, it cannot cope with the massive labeling environment with singularity and hysteresis. AMCT is introduced to optimize it, i.e., CT-based AMCT algorithm. When reading many RFID tags during multi-tag identification, there may be a collision problem. This can be effectively solved with the AMCT algorithm. It modifies the tree’s bifurcation dynamically to improve recognition performance. The method uses a tree structure to organize the label identification process in situations where labels are not recognized. A branch indicates a different recognition path, and each node represents a possible label recognition state. After the reader sends a query signal to the tag and detects a collision, the algorithm adjusts the number of branches on the tree based on the current collision situation. If there are fewer collisions, fewer forks may be used. However, if there are more collisions, the number of forks is increased. The algorithm then recursively processes each fork to identify each tag individually. Each fork handles a smaller set of tags. This method dynamically adjusts the bifurcation number to quickly adapt to the current collision situation and reduce of unnecessary queries. The bifurcation number is automatically adjusted based on label density and environmental conditions, enhancing its versatility for various applications. Adaptive Collision Tree (ACT) incorporates a collision factor, i.e., depending on the actual situation of the collision, it determines whether the bifurcation method is binary or quadratic in order to comply with the condition of multi-tag volume (Farah, Chabir & Abdelkrim, 2023). However, this causes the amount of free time slots in the model to rise and further optimization is required. The AMCT algorithm solves the problem of the amount of free time slots that exists in the ACT algorithm by identifying colliding labels and then adaptively selecting binary and octal forks. Equation (4) displays the identification probability of labels in adaptive forking. $p (k) = [1 + (1 / L - 1)^{N - 1}]^{k - 1} * (- 1 / L + 1)^{N - 1}$ (4)

In Equation (1), L is the number of forks and k is the search depth. Therefore, the expectation of the mean value of search depth in recognizing labels in a multinomial tree is shown in Equation (5). $E (k) = \sum_{k = 1}^{\infty} kp (k) = \sum_{k = 1}^{\infty} k {(1 - 1 / L)}^{N - 1} * {[1 - {(1 - 1 / L)}^{N - 1}]}^{k - 1}$ (5)

The mean value of the number of time slots T required by the system when the number of recognized tags is N can be obtained by further arithmetic summation as shown in Equation (6). $T = L * E (k) = \frac{1}{{(1 - 1 / L)}^{N - 1}}$ (6)

The system recognition accuracy will increase as the forked tree grows, but it will also increase the computational load on the model. In this scenario, the use of the adaptive forking CT algorithm is most appropriate, as the mean value of the number of time slots is at its lowest when the number of forks equals the total number of tags. The number of binary and octree trees are substituted into Equation (6) to obtain the corresponding T-value Equation, and the comparison shows that when the total number of labels ⩾4, octree is required, and vice versa, binary tree is selected. A single octree bifurcation may make some groups empty and produce unnecessary waste of resources, so the concept of prefix prediction is introduced to predict the already existing prefix sequences, as Fig. 4 depicts the details.

Fig. 4

Prefix sequence prediction process.

After the RFID system recognizes the collision information, the reader-writer issues the prefix prediction command. This directs all tags in the collision time slot to extract the values of the first three collision bits (a1/a2/a3), which are then converted into decimal. The system also returns a fresh binary sequence to the reader-writer after initializing it. With this additional data, the algorithm is able to locate the collision bits, translate them back into binary, and finally produce a new query prefix (Hidayat, Ali & Arshad, 2022). The AMCT algorithm also introduces the concept of automatic identification, when the collision bit is one, the reader/writer automatically sets 0/1 to the corresponding bit, eliminating the need for unnecessary query steps. Figure 5 depicts the AMCT algorithm’s entire operating flow.

Fig. 5

Overall operation flow of AMCT algorithm.

The reader must first read the top-of-stack prefix and issue a search query to ask all tags to reply. This may be broadly broken down into five steps. Then the RFID tags that meet the conditions will feedback and deliver the remainder to the reader/writer. The reader-writer uses the tag response situation to determine the time slot. If there are many tags, it is considered a collision time slot. In such cases, the reader-writer will select a multinomial tree search based on the actual number of colliding tags and choose a new query prefix simultaneously (Liu et al., 2020). The above steps are repeated until the query stack is empty to ensure that all RFID tags are recognized, and the recognition tree structure is shown in Fig. 6.

Fig. 6

RFID tag identification tree structure.

The key metrics affecting the collision avoidance performance of an RFID system include recognition time, throughput rate and communication data volume. Assume that the RFID range contains unidentified tags as m and the length of the tag ID sequence is d bits. It is known that there are only two kinds of CT nodes, leaf nodes and non-leaf nodes, the former, i.e., unidentified tags m, and the latter are collision nodes, which are divided into octal collision nodes C_8ary (m) and binary collision nodes C_2ary (m). Combining with the concept of automatic identification, the total number of time slots T_AMCT (m) of the algorithm can be obtained as shown in Equation (7) (Andresini, Appice & Malerba, 2021). $T_{AMCT} (m) = 2 C_{8 ary} (m) + C_{2 ary} (m) + m - 2 M_{1}$ (7)

In Equation (7), M₁ is the number of one-bit collisions.

The further crucial indicator for assessing its success is the throughput rate, which goes hand in hand with the number of time slots (Alotaibi et al., 2023). Equation (8), which displays the ratio of the total number of time slots needed to the number of unidentified tags, represents the system throughput rate (Sarkar et al., 2022). $S_{AMCT} = \frac{m}{T_{AMCT} (m)}$ (8)

There is also a strong correlation between system performance and the complexity of the communication, when the communication is simpler, the system works naturally more efficiently, and this metric is often judged by the amount of reader-tag communication. The amount of interactive communication is shown in Equation (9) (Lalbakhsh et al., 2023). $L_{AMCT} (m) = \sum_{i}^{T_{AMCT} (m) - C_{8 ary} (m)} (l_{com} - l_{ID}) + C_{8 ary} (m) * (8 + d / 2)$ (9)

In Equation (9), l_com denotes the value of the command information length delivered by the reader when querying the command, which takes the value of the interval [1,d], so the mean value d/2 is selected. l_ID denotes the value of the information length fed back by the tag when querying the command, which is 8 bits. The unit of information length is bit. In subsequent applications, the number of electronic tags is in the range [1,2000], the tag recognition speed is not less than 50tag/s, and the channel frequency is 806 MHz.

3 Result and discussion

To validate the effectiveness of the AMCTRFID based technique proposed by the study, the study conducted performance simulation experiments on it. Firstly, the modules of AMCT algorithm such as adaptive and multinomial tree were verified to ensure that the optimization is effective. Then the performance of the AMCT algorithm is analyzed with the rest of the anti-collision algorithms and its application in RFID-NID technology.

3.1 AMCT-RFID model validation and performance

Based on the collision prevention algorithm’s performance indicators, such as the accuracy rate, time slot count, and system throughput rate. The study examines and validates the algorithm’s performance. Table 1 displays the experimental setup as well as the default parameter values.

Table 1
Overall experimental environment and model parameter Settings

Name Environment/Parameters

Operating system Windows 10

CPU i7-10700

GPU GTX 2070 Super

Internal memory 32G

Development environment Spyder 3.3.6 (Python 3.6)

Number of readers 1

Label serial number length 95 bit

Label number change range [100,2000]

Data transfer rate 100kbps

Name	Environment/Parameters
Operating system	Windows 10
CPU	i7-10700
GPU	GTX 2070 Super
Internal memory	32G
Development environment	Spyder 3.3.6 (Python 3.6)
Number of readers	1
Label serial number length	95 bit
Label number change range	[100,2000]
Data transfer rate	100kbps

The data transfer rate indicates the data interaction transfer rate between both the reader and the RFID tag. Firstly, the study analyzed the performance of the adaptive module of the algorithm, and compared the ACT with the initial CT algorithm, using the recognition probability of NI and the system throughput as the judging indexes, and the experimental results are shown in Fig. 7.

Fig. 7

Experimental results of adaptive module optimization performance in anti-collision algorithm.

In Fig. 7(a), the ACT algorithm with the introduction of the adaptive module performs significantly better in the interval of the number of labels [0,5]. When there are 0.5 labels, the CT algorithm reaches its maximum value of 18%, whereas the ACT algorithm reaches its maximum value of 37% when there are 1 labels, or roughly twice the peak of the CT algorithm’s throughput rate. At the same time, the CT algorithm’s throughput rate drops to its lowest value at 0.35 labels, while the ACT algorithm’s throughput rate still reaches about 14%. The recognition accuracy of the ACT method for NI is always higher than that of the CT algorithm in Fig. 7(b), where the number of tags is 200/400/600/800, respectively. This difference in recognition accuracy between the two algorithms is even bigger as the number of tags increases. When the labels reaches 200, the ACT/CT algorithm’s accuracy is 92.47% and 89.14%, respectively, exhibiting a difference of only 3.33%. However, when the labels increases to 800, the ACT/CT algorithm’s accuracy drops to 79.46% and 67.96%, respectively, indicating a difference of 11.5%. The mean recognition accuracy of the ACT algorithm is 86.92%, and the mean recognition accuracy of the CT algorithm is 78.22%. Although the performance of the ACT algorithm has a certain degree of improvement relative to the original algorithm, there is still room for improvement, which also side-steps the need for optimization of the multinomial tree. The performance of the AMCT algorithm under different amount of labels is shown in Fig. 8.

Fig. 8

Experimental results of AMCT algorithm performance under different label quantities.

Figure 8(a) above shows the change curve of system throughput rate generated by different collision labels under different label numbers, and when M = 4, it is the key point where the system performance is transformed, which also proves that the model is right in choosing this data point as the automation selection threshold of the multinomial tree. When M < 4, the system throughput rate has not been completely increased, and the average throughput rate is 57%, while when M > 4, the system throughput rate gradually decreases, and when the value of M rises to 7, the throughput rate of the model decreases to 56%, which is a reduction of 6% compared to M = 4. And the throughput rate at that point fluctuates little and is more stable. Figure 8(b) illustrates the system’s time slot requirements for different collision labels. The figure shows that the number of required time slots increases as the number of labels increases, with values greater than 4. Moreover, collision label 4 requires 700 fewer time slots than collision label 2 when the total number of labels is 2000. In conclusion, the study of selecting a multinomial tree as well as the selection of the critical point is necessary and correct.

3.2 RFID Technology in NID with AMCT algorithm

The study introduces CT algorithm, ACT algorithm, An Adaptive Assigned Tree Slotted AlohaAnti-collision Protocol (AdATSA) algorithm, and Optimal Binary Tracking Tree (OBTT) algorithm, which have been proposed AMCT algorithms to be compared and the comparison results in the number of time slots metric are shown in Fig. 9.

Fig. 9

Comparison of the number of time slots of different algorithms.

Figure 9(a) presents the curve depicting the variation in the number of collision time slots for each algorithm in relation to the total number of tags. It is evident that the CT algorithm which was not optimized displays the poorest performance, with a final number of collision time slots reaching 1,928. Following adaptive improvement, the ACT algorithm exhibits a certain degree of optimization. The AMCT algorithm proposed in the study has a clear advantage due to its slow rise. There is minimal difference in the performance of AdATSA and OBTT algorithms, with collision time slots falling within the range of around 1000. Moreover, the collision time slots are reduced by 428. Compared to the remaining four algorithms, the AMCT algorithm exhibits a significant reduction of 67.12% in the final collision time slots, which is 812. Figure 9(b) displays the variation curve of the total time slots required for each algorithm, based on the given condition of 2000 total tags. The study found that the AMCT algorithm requires 2853 time slots, whereas the CT algorithm, ACT algorithm, AdATSA algorithm, and OBTT algorithm require 3916, 3507, 3044, and 3167 time slots respectively. Consequently, the AMCT algorithm exhibits an average reduction of 19 time slots. 47% of the total number of time slots was taken up by this algorithm, which is higher than the other four algorithms. The variation in throughput for each algorithm is displayed in Fig. 10.

Fig. 10

Comparison of throughput performance of each algorithm system.

Figure 10(a) illustrates the system’s throughput rate for each algorithm. The AMCT and AdATSA algorithms demonstrate relative stability, while the other algorithms show significant fluctuation. The AMCT algorithm proposed in this study has the highest throughput rate with an average of 62.1%, and both the AdATSA and OBTT algorithms also achieve over 60% accuracy, while all CT algorithms fall below 60%. The ACT algorithm shows a 3.8% improvement over the CT algorithm, whereas the AMCT algorithm demonstrates an 18.9% improvement over the CT algorithm. Figure 10(b) shows the algorithmic throughput rates, with the CT algorithm having the highest rate and the AMCT algorithm ranking second. The AdATSA algorithm has a similar throughput rate to the AMCT algorithm, but its stability is significantly worse. The OBTT algorithm is only surpassed by the AdATSA algorithm, and together they account for over 60% of the total fluctuations. Further applying each algorithm to the NI experiments, the model’s own performance and intrusion detection performance are shown in Table 2.

Table 2

Compares the performance of each algorithm in label recognition and intrusion recognition

Index	CT	ACT	OBTT	AdATSA	AMCT
Model throughput (%)	52.2%	56.0%	60.1%	60.3%	62.1%
Throughput gap with AMCT (%)	18.9%	10.9%	3.3%	2.8%	/
Fluctuation degree *102	96	80	84	137	39
Intrusion identification accuracy (%)	86.92%	78.22%	93.67%	95.41%	99.82%
F1 (%)	85.41%	78.09%	93.54%	84.23%	99.41%

According to Table 2, the AMCT algorithm achieves the highest system throughput, exceeding the remaining four algorithms by an average of 8.98%. While there are substantial differences in the degree of fluctuation amongst the algorithms, excluding the AdATSA algorithm, the degree of fluctuation for the remaining algorithms falls below 10,000. Notably, the degree of fluctuation for the AMCT algorithm is reduced by an average of 53.8% compared to the rest of the algorithms. The intrusion detection accuracy of the AdATSA, OBTT, and AMCT algorithms exceeds 90%. On average, the AMCT algorithm outperforms the others by 11.265%, and the false alarm rate follows the same pattern, with the AMCT algorithm surpassing the others by an average of 12.17%. In conclusion, the aforementioned data indicates that the research, which proposes RFID technology employing the AMCT algorithm, exhibits superior performance in NID. The introduction of multi-tree and adaptive modules into the AMCT algorithm results in a recognition performance that is more stable under various tag numbers and is not affected by an increase in tag numbers. This also reduces retrieval times and contributes to improved recognition accuracy. Therefore, the proposed algorithm enhances both label recognition performance and efficiency. The study compares the communication complexity and efficiency change curves of each algorithm model in the applied process. The obtained experimental results are presented in Fig. 11.

Fig. 11

Visual analysis of the actual application effect of each model.

Based on Fig. 11(a), it can be observed that the communication complexity of all models increases as the number of labels increases. The OBTT model exhibits the highest rate of increase and ultimately achieves a final communication complexity of 81.1*105, which is 50.31% greater than that of the research design algorithm. The average communication complexities of the OBTT, AdATSA, and research design algorithms are 51.4*105, 28.5*105, and 19.7*105, respectively. Figure 11 (b) shows that the research design algorithm has an average communication complexity 61.67% lower than that of the OBTT algorithm and AdATSA algorithm. The efficiency of each model remains stable, and the research design algorithm has an average efficiency of 78.29%, surpassing the OBTT algorithm and AdATSA algorithm by 15.98% and 8.24%, respectively. In conclusion, the study’s proposed multi-label anti-collision algorithm has produced superior results in practical applications. It has also significantly reduced the model’s communication complexity and improved operating efficiency.

4 Conclusion

Finally, this paper presents and validates an algorithm for achieving efficient NID in high-traffic networks based on RFID. The paper’s main contributions include designing an RFID-NID algorithm based on the CT anti-collision algorithm to detect large tag responses and introducing an adaptive multi-tree to enable the adaptive processing mode of tag interference. It serves as a guide for the future implementation of RFID and network technologies and is well-suited for the current big data landscape. The outcomes indicated a 14% enhancement in the throughput rate and an 8.7% rise in intrusion recognition accuracy upon incorporating the adaptive module. Furthermore, the selection of the multinomial tree critical point led to a 6% improvement in the throughput rate. The results showed that the adaptive tree algorithm, which uses collision counting binary or octal segmentation, effectively addresses RFID tag collision and improves throughput. Furthermore, when integrated with NID, the algorithm achieved a threat identification accuracy of 99.82%, surpassing current technologies. The experiment also revealed an 8.98% increase in throughput, a 67.12% decrease in collision slots, and an average fluctuation reduction of 53.8%.

The experimental results demonstrate the benefits of the AMCT algorithm in preventing collisions and identifying NIs in multi-label RFID systems. The algorithm substantially enhances the throughput rate and the accuracy of intrusion recognition by integrating adaptive modules and multi-tree technology. This indicates that the algorithm can adapt to real-time environments, enhancing its ability to handle complicated or shifting label response scenarios. The AMCT algorithm reduces the number of collision slots and total collisions, indicating superior efficiency in collision management in high-density label environments. The AMCT algorithm demonstrates high accuracy in recognizing NIs, as evidenced by its F1 value. Its efficient and precise recognition ability is critical for network security, particularly in the presence of vast quantities of data and complex network environments. Additionally, the algorithm exhibits low fluctuations, resulting in relatively stable performance under varying conditions, which is crucial for practical deployment. The AMCT algorithm offers significant benefits in enhancing the effectiveness of multi-label RFID systems, minimizing collisions, and improving the precision of NID. These benefits provide substantial technical support for the implementation of RFID technology and the field of network security.

However, the study only improved the adaptive changes in labels. While the model’s generalization ability has improved to some extent, it still takes into account the label movement. Label movement can cause them to enter or leave the reader’s reading area, resulting in lost label signals or sudden appearance of new ones. Simultaneous movement of multiple tags can cause read conflicts, as the response signals of the tags may overlap and interfere with the reader’s ability to accurately identify individual tag information. Therefore, future research should focus on designing anti-collision algorithms that can adapt to the dynamic movement of tags. Different methods were employed to reduce reading errors in the label movement environment, optimizing the overall system’s flexibility and adaptability.

References

Abbasian,

, Safkhani,

(2020) CNCAA: A New Anti-collision Algorithm using both Collided and Non-collided Parts of Information. Computer Networks, 172(3), 107159–107167.

Ai,

, Bai,

, Xu,

, Zhang,

(2022) Anti-collision algorithm based on slotted random regressive-style binary search tree in RFID technology. IET Communications, 16(10), 1200–1208.

Alotaibi,

, Murad,

, Tasadduq,

I. A.

, Alhuthali,

S. A. H.

, Al-Osaimi,

F. R.

, Aldosari,

, Rushdi,

A. M.

(2023) Anti-collision algorithm for identification in precision agriculture applications. IEEE Access, 11(1), 130197–130205.

Andresini,

, Appice,

, Malerba,

(2021) Autoencoder-based Deep Metric Learning for Network Intrusion Detection. Information Sciences, 569(1), 706–727.

Baygin,

, Yaman,

, Baygin,

, MK,

(2022) A blockchain-based approach to smart cargo transportation using UHF RFID. Expert Systems with Applications, 188(1), 116030–116030.

Chen,

, Luo,

, Xiang,

, Chen,

, Truong,

H. L.

(2021) IoT Cloud Security Review: A Case Study Approach Using Emerging Consumer-oriented Applications. ACM Computing Surveys, 54(4), 75–111.

Chen,

, Yang,

, Liu,

, Xu,

, Luo,

, Zhang,

(2022) Efficient and accurate identification of missing tags for large-scale dynamic RFID systems. Journal of Systems Architecture, 124(1), 102394–102404.

Dutang,

, Guibert,

(2022) An explicit split point procedure in model-based trees allowing for a quick fitting of GLM trees and GLM forests. Statistics and Computing, 32(1), 10059–10064.

Farah,

, Chabir,

, Abdelkrim,

M. N.

(2023) High level Petri nets-based proposal of an integrated intrusion detection and prevention mechanism in network controlled systems. IET Communications, 17(4), 469–477.

10.

Ghasemi,

, Babaie,

(2022) A lightweight secure authentication approach based on stream ciphering for RFID-based Internet of Things. Computers and Electrical Engineering, 102(1), 108288–108294.

11.

Hidayat,

, Ali,

M. Z.

, Arshad,

(2022) Machine learning-based intrusion detection system: an experimental comparison. Journal of Computational and Cognitive Engineering, 2(2), 88–97.

12.

Hu,

, Zhang,

, Wang,

, Zhao,

, Wang,

(2020) An RFID Anti-collision Algorithm Based on Equal Area Division. Journal of Jilin University(Science Edition), 58(1), 120–126.

13.

Kan,

, Fan,

, Fang,

, Cao,

, Li,

(2021) A Novel IoT Network Intrusion Detection Approach Based on Adaptive Particle Swarm Optimization Convolutional Neural Network. Information Sciences, 568(1), 147–162.

14.

Lalbakhsh,

, Yahya,

S. I.

, Moloudian,

, Hazzazi,

, Sobhani,

S. N.

, Assaad,

, Chaudhary,

M. A.

(2023) Hybrid encoding method for radio frequency identification in the internet of things systems. IEEE Access, 11(1), 122554–122565.

15.

Li,

, Sun,

, Li,

, Wu,

, Inserra,

, Su,

, Fang,

, Wen,

(2022) A Dynamic Multi-ary Query Tree Protocol for Passive RFID Anti-collision. Computers, Materials and Continuum, 000(9), 4931–4944.

16.

Liu,

, Wen,

, Sun,

, Zhang,

(2020) A novel multiway splits decision tree for multiple types of data. Mathematical Problems in Engineering, 1(3), 7870534–7870546.

17.

Martin,

M. L.

, Carro,

, Arribas,

, Esguevillas,

A. S.

(2021) Network intrusion detection with a novel hierarchy of distances between embeddings of hash IP addresses. Knowledge-based Systems, 219(3), 106887–106904.

18.

Pan,

, Miao,

(2023) Dynamics and risk assessment of a remanufacturing closed-loop supply chain system using the internet of things and neural network approach. The Journal of Supercomputing, 79(4), 3878–3901.

19.

Radi,

, Oli,

, Kiljo,

(2020) Anticollision algorithm for radio frequency identification system with low memory requirements. Transactions on Emerging Telecommunications Technologies, 31(12), 3969–3983.

20.

Saghlatoon,

, Mirzavand,

, Mousavi,

(2020) Fixed-frequency low-loss dielectric material sensing transmitter. IEEE Transactions on Industrial Electronics, 68(4), 3517–3526.

21.

Sarkar,

, Takeyeva,

, Guchhait,

, Sarkar,

(2022) Optimized radio-frequency identification system for different warehouse shapes. Knowledge-Based Systems, 258(9), 109811–109823.

22.

Samsami,

M. M.

, Yasrebi,

(2021) Novel RFID anti-collision algorithm based on the Monte-Carlo query tree search. Wireless Networks, 27(1), 621–634.

23.

Srivastava,

, Gupta,

, Quamara,

, Chaudhary,

, Aski

VJ.

(2020) Future IoT-enabled threats and vulnerabilities: State of the art, challenges, and future prospects. International Journal of Communication Systems, 33(12), 4443–4483.

24.

Wang,

, Shao,

, Meng,

, Gao,

(2021) A Physical Layer Network Coding Based Tag Anti-Collision Algorithm for RFID System. Computers, Materials and Continuum, 000(001), 931–945.

25.

Zhao,

, Xiong,

, Wang,

, Chen,

, Xu,

(2022) Wear-free gesture recognition based on residual features of RFID signals. Intelligent Data Analysis, 26(4), 1051–1070.

26.

Zhang,

, Hu,

, Yan,

, Yang,

(2020) Structure optimization of an ultrahigh frequency radio frequency identification tag thread based on the normal mode helix dipole antenna. Textile Research Journal, 91(5), 523–532.

27.

Zhou,

, Jiang,

(2021) Research on hybrid of ALOHA and multi-fork tree Anti-collision algorithm for RFID. Procedia Computer Science, 183(5), 389–394.