Method and system for detecting security of Chinese character click verification code based on deep learning

Abstract

The invention discloses a method and a system for detecting the security of a Chinese character click verification code based on deep learning. The system comprises a client device, a distributed database device, a telescopic server, a cloud platform device, a detection system processing device, a real-time monitoring device and a terminal visualization large screen. The method comprises the following steps: marking Chinese character positions of Chinese character click verification code data, and training a Chinese character position detection model constructed based on a YOLOv5 target detection algorithm according to an obtained Chinese character position detection data set; and training a Faster-RCNN network through a Chinese character click verification code data set detected by the trained Chinese character position detection model so as to perform Chinese character click verification code recognition and obtain a verification code recognition result. According to the method, the YOLOv5 target detection algorithm and the Faster-RCNN network are utilized, so that the positions and categories of the Chinese characters of the Chinese character verification code can be accurately recognized, the error recognition rate is reduced, and the overall recognition efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and pattern recognition, and in particular to a method and system for detecting the security of a Chinese character point-selection verification code based on deep learning. Background Art

[0002] Verification code is a public automatic recognition program. The verification code image is often composed of numbers, letters and some interference items, which is used to determine whether the service object is a human or an automatic program. In order to improve the anti-attack and network security of the website, the verification code technology has gradually developed from the recognition of numbers and letters to the recognition of Chinese character point verification code, which undoubtedly increases the difficulty of verification code recognition; at the same time, compared with automatic programs, although humans are more likely to recognize the information in Chinese character verification code images, with the development of artificial intelligence technology, malicious automatic programs are also constantly updated and iterated, trying to circumvent various verification code recognition mechanisms (Hu Zhengbing, Lu Hong, Liao Longlong. Computer and Modernization, 2011, 1(2):1). Therefore, in order to prevent the service from being used arbitrarily, it is of great significance to find the design loopholes of verification code recognition, identify the website text verification code images, and reduce the threat of network information.

[0003] Currently, in the field of Chinese character point verification code recognition, image matching is mainly used to match and recognize image text and text converted into images. This method has limited application scenarios and cannot be applied to all Chinese characters, such as deformed Chinese character images, which leads to low efficiency in Chinese character point verification code recognition (Zhang Jingwen, Lin Pingzhi, Xiao Sijia, etc. Advances in Applied Mathematics, 2021, 10:4006).

[0004] In view of the above problems, the present invention proposes a method and system for detecting the security of Chinese character point-selection verification codes based on deep learning, which can realize interference-resistant Chinese character point-selection verification code recognition. Summary of the invention

[0005] The main purpose of the present invention is to provide a method and system for detecting the security of Chinese character point verification codes based on deep learning, aiming to solve the technical problem that the existing methods are inefficient in recognizing Chinese character verification codes.

[0006] By using the YOLOv5 target detection algorithm and the Faster-RCNN network, the detection method and system for the security of Chinese character point-selection verification codes based on deep learning of the present invention can effectively improve the security and recognition efficiency of Chinese character point-selection verification codes, prevent automated attacks, and ensure the security of network services.

[0007] To achieve the above purpose, the technical solution provided by the present invention is as follows:

[0008] Step 1: System construction.

[0009] The Chinese character verification code security detection system based on deep learning is composed of the following key modules, including: client equipment, distributed database equipment, scalable server, cloud platform equipment, detection system processing equipment, real-time monitoring equipment and terminal visualization large screen;

[0010] The client device is connected to the distributed database device, and the distributed database device is connected to the telescopic server, the cloud platform device, the detection system processing device, the real-time monitoring device and the terminal visualization large screen in sequence.

[0011] The client device is a terminal device through which the user interacts with the verification code. It is used to send a verification code request to the server and receive a verification code image, as well as to feedback the user's clicking behavior (i.e. the selected Chinese character position) to the back-end server; it is also responsible for sending the recognition result to the back-end server for processing.

[0012] The distributed database device is used to store data required by the system, including user information, verification code data sets, model training data and verification code recognition results; the distributed database device supports the access of large-scale data through distributed storage technology, ensuring high availability and high concurrent processing capabilities of the data.

[0013] The retractable server, as the computing engine of the system, is responsible for processing the verification code request from the client device. The retractable server supports dynamic expansion and can adjust the allocation of computing resources according to the real-time load situation. It is used to receive the verification code request from the client device and distribute the verification code request to the cloud platform device, and exchange data with other devices to dynamically adjust the allocation of computing resources.

[0014] The cloud platform device is used to integrate a deep learning model (such as a Chinese character position detection model based on the YOLOv5 target detection algorithm and a Faster-RCNN network) to perform image processing and analysis on the verification code images transmitted to the cloud. The cloud platform device not only supports the rapid recognition of verification code images, but also can perform training and reasoning tasks, and feed back the verification code recognition results uploaded by the cloud platform device to the client device in real time.

[0015] The detection system processing device is responsible for the real-time recognition and security detection of the Chinese character verification code, and is used to perform position detection and classification recognition on the verification code image through the Chinese character position detection model and the Faster-RCNN network, obtain the verification code recognition result, and upload it to the cloud platform device, and then upload it to the telescopic server or client device through the cloud platform device.

[0016] The real-time monitoring device is used to monitor the system status, which includes the verification code recognition process, cloud platform resource usage and data transmission process; the real-time monitoring device can alarm in real time and automatically adjust the system resource configuration to ensure the stability and efficient operation of the system.

[0017] The terminal visualization screen is responsible for displaying the system operation status and test results, and is used to display the system operation status, verification code recognition results, and the administrator operation interface. Through the visualization administrator operation interface, the administrator can view the working status of each system module, the number of verification codes processed, recognition accuracy, and other information in real time, and perform system management and optimization.

[0018] Step 2: Click on the Chinese character verification code to recognize it.

[0019] The above-mentioned Chinese character point verification code security detection system based on deep learning is implemented by a Chinese character point verification code security detection method based on deep learning, and the method is specifically as follows:

[0020] Acquire Chinese character click verification code data, and mark the Chinese character positions in the Chinese character click verification code data to obtain a Chinese character position detection data set;

[0021] Training a preset Chinese character position detection model according to the Chinese character position detection data set, and using the training result for the Chinese character selection task, wherein the Chinese character position detection model is constructed based on the YOLOv5 target detection algorithm;

[0022] The trained Chinese character position detection model is used to detect Chinese characters on the Chinese character click verification code images, and each Chinese character rectangular frame is saved as a single Chinese character image. At the same time, the Chinese characters corresponding to the Chinese character images are marked to obtain the Chinese character click verification code dataset;

[0023] A preset Faster-RCNN network is trained according to the Chinese character click verification code data set to obtain a Chinese character click verification code network model;

[0024] Based on the Chinese character selection verification code network model, Chinese character selection verification code recognition is performed to obtain a verification code recognition result.

[0025] Step 3: Specific steps for Chinese character position detection.

[0026] The step of performing Chinese character detection on the Chinese character point verification code picture by using the trained Chinese character position detection model to obtain a Chinese character point verification code data set includes:

[0027] Perform Chinese character detection on the Chinese character point verification code picture by using the trained Chinese character position detection model to obtain the Chinese character position and position coordinates in the Chinese character point verification code picture;

[0028] According to the position coordinates, a Chinese character rectangular frame corresponding to each Chinese character position is cut out from the Chinese character point selection verification code picture, and the Chinese character rectangular frame is converted into a Chinese character picture;

[0029] Mark the Chinese characters corresponding to the Chinese character images and generate corresponding txt files;

[0030] The Chinese character images and the txt files are collected, and the file names of the Chinese character images are matched with the file names of the txt files to generate a Chinese character point verification code data set.

[0031] Step 4: Specific steps for training the preset Faster-RCNN network.

[0032] The step of training a preset Faster-RCNN network according to the Chinese character click verification code data set to obtain a Chinese character click verification code network model includes:

[0033] Resampling a plurality of Chinese character images in the Chinese character click verification code data set to obtain resampled images of the same size as the Chinese character images;

[0034] The resampled image and the txt file in the Chinese character click verification code data set are input into the preset Faster-RCNN network for training to obtain the Chinese character click verification code network model.

[0035] Step 4: Specific steps for the Chinese character click verification code network model to recognize text.

[0036] The step of performing Chinese character click verification code recognition based on the Chinese character click verification code network model to obtain a verification code recognition result comprises:

[0037] First, a Chinese character point-selection verification code is input into a deep convolutional network for feature extraction to obtain deep features. The Faster-RCNN network comprises a deep convolutional network and a region candidate network. The deep convolutional network comprises a convolutional layer, a ReLU activation layer, and a spatial region pyramid pooling layer. The deep convolutional network of the Faster-RCNN network comprises 13 convolutional layers, 13 ReLU activation layers, and 4 spatial region pyramid pooling layers. The deep convolutional network is used to extract a feature map of an input image. The Faster-RCNN network classifies the feature map through a fully connected layer and outputs a verification code recognition result.

[0038] Then, the candidate box and regional candidate coordinates corresponding to the Chinese character point selection verification code are determined through the regional candidate network, and the regional candidate network consists of candidate box classification and candidate box position correction; wherein, the candidate box classification realizes whether all preset candidate boxes are Chinese characters through a convolution layer, two reshape layers and a softmax layer; the candidate box position correction calculates the offset of the candidate box through a convolution layer to obtain accurate regional candidate coordinates.

[0039] For accurate classification, the depth feature and the region candidate coordinates are input into the spatial region pyramid pooling layer to generate the region candidate depth feature;

[0040] Finally, the candidate deep features of the region are classified according to the overlapping convolutional layer and the ReLU activation layer to obtain the verification code recognition result of the Chinese character point selection verification code, including the precise position and classification result.

[0041] One or more technical solutions proposed by the present invention have at least the following technical effects: The present invention uses the YOLOv5 target detection algorithm and the Faster-RCNN network to accurately identify the position and category of Chinese characters, reduce the misrecognition rate, improve the overall recognition efficiency, and achieve high-accuracy anti-interference Chinese character point verification code recognition. The invention can be widely used in various network platforms that need to verify the identity of users, can improve the security level of verification codes, prevent attacks from automatic programs, and ensure the security of network services, thereby improving the reliability of data transaction Chinese character point verification user identity authentication and the security of data transactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 This is a flow chart of the point-to-point verification code recognition based on the Faster-RCNN network of the present invention;

[0045] Figure 2 This is an example diagram of the click verification code provided by the present invention;

[0046] Figure 3 This is a structural schematic diagram of the detection system for the security of Chinese character point-selection verification codes based on deep learning of the present invention.

[0047] Description of Figure Numbers:

[0048] 1. Client device; 2. Distributed database device; 3. Scalable server; 4. Detection system processing equipment; 5. Real-time monitoring equipment; 6. Terminal visualization large screen.

[0049] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0050] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present invention and are not used to limit the present invention.

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, close, far...) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0053] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0054] The embodiment of the present invention provides a method for detecting the security of Chinese character point-selection verification codes based on deep learning. In this embodiment, the method for detecting the security of Chinese character point-selection verification codes based on deep learning includes the steps of: obtaining Chinese character point-selection verification code data, and marking the Chinese character positions in the Chinese character point-selection verification code data to obtain a Chinese character position detection data set; training a preset Chinese character position detection model based on the Chinese character position detection data set, and the Chinese character position detection model is constructed based on the YOLOv5 target detection algorithm; performing Chinese character detection on a Chinese character point-selection verification code image through the trained Chinese character position detection model to obtain a Chinese character point-selection verification code data set; training a preset Faster-RCNN network based on the Chinese character point-selection verification code data set to obtain a Chinese character point-selection verification code network model; performing Chinese character point-selection verification code recognition based on the Chinese character point-selection verification code network model to obtain a verification code recognition result.

[0055] It should be noted that the Chinese character click verification code data may be data involved in allowing users to perform click verification. The Chinese character click verification code allows users to click and select the correct Chinese characters or related content from some given Chinese characters according to prompts to complete the verification, which is used to ensure that it is the user who is performing the operation to enhance security and accuracy.

[0056] It is understandable that the YOLOv5 target detection algorithm is a real-time target detection algorithm that can directly divide the grid on the image and predict the category and bounding box information of the target in each grid. Through the above-mentioned Chinese character position detection dataset with the Chinese character positions annotated, the model built based on the YOLOv5 target detection algorithm can be trained to obtain a Chinese character position detection model that can classify Chinese character point verification code images and is used for Chinese character box selection tasks.

[0057] It should be understood that the Chinese character click verification code data set is a data set obtained by classifying and converting each Chinese character in the Chinese character click verification code image into an image through a Chinese character position detection model. For example, a trained Chinese character position detection model can be used to perform Chinese character detection on the Chinese character click verification code image, and the rectangular frame of each Chinese character can be saved as a single image, and the Chinese character corresponding to the image can be annotated to form a Chinese character click verification code data set.

[0058] It should be noted that the Faster-RCNN (Faster Region-based Convolutional Neural Network) network is a target detection network based on a deep learning algorithm for detecting and identifying target objects in images. The Faster-RCNN network can be composed of a series of convolutional layers, a region proposal network (RPN), and a classification and regression layer, which can quickly and accurately extract the target region in the image, and classify and locate the target.

[0059] Through the above classified Chinese character click verification code data set, the Faster-RCNN network can be trained for deep learning to obtain the Chinese character click verification code network model.

[0060] In this implementation, by using the YOLOv5 target detection algorithm and the Faster-RCNN network, the security and recognition efficiency of the Chinese character point verification code can be effectively improved, automated attacks can be prevented, the security of network services can be guaranteed, and the problem of low efficiency of the existing method in recognizing Chinese character verification codes can be solved.

[0061] In a feasible implementation manner, the step of performing Chinese character detection on a Chinese character click verification code picture through a trained Chinese character position detection model to obtain a Chinese character click verification code data set described in this embodiment includes: performing Chinese character detection on a Chinese character click verification code picture through a trained Chinese character position detection model to obtain the Chinese character position and position coordinates in the Chinese character click verification code picture; according to the position coordinates, cutting out a Chinese character rectangular frame corresponding to each Chinese character position from the Chinese character click verification code picture, and converting the Chinese character rectangular frame into a Chinese character picture; marking the Chinese characters corresponding to the Chinese character picture to generate a corresponding txt file; collecting the Chinese character picture and the txt file, and making the file name of the Chinese character picture correspond to the file name of the txt to generate a Chinese character click verification code data set.

[0062] It should be noted that the txt file can be a file used to store the plain text information of the Chinese characters corresponding to the Chinese character images.

[0063] In this implementation, a trained Chinese character position detection model is used to detect all Chinese character positions in a Chinese character verification code image, and the position coordinates are returned; then, based on the returned Chinese character position coordinates, a Chinese character rectangular frame corresponding to each Chinese character position is cropped from the Chinese character point verification code image and saved as an image; then, the Chinese character image is annotated as Chinese characters, and a txt file is generated; finally, the above-obtained images and txt files are collected, and the Chinese character image file names are matched one by one with the txt file names to form a Chinese character point verification code data set.

[0064] In a feasible implementation manner, the step of training a preset Faster-RCNN network according to the Chinese character click verification code data set to obtain a Chinese character click verification code network model described in this embodiment includes: resampling multiple Chinese character pictures in the Chinese character click verification code data set to obtain resampled images of the same size as the Chinese character pictures; inputting the resampled images and the txt files in the Chinese character click verification code data set into the preset Faster-RCNN network for training to obtain a Chinese character click verification code network model.

[0065] It should be noted that the resampled image may be an image obtained by resampling the pixels of the Chinese character image. By resampling, the number of pixels of the Chinese character image may be increased, for example, by upgrading a low-resolution image to a higher resolution (up-sampling), so as to maintain the characteristics and quality of the image as much as possible.

[0066] In this implementation, the Faster-RCNN network is trained by using the resampled image obtained after resampling and the txt file in the Chinese character point verification code data set, which can further improve the accuracy of the Chinese character point verification code network model recognition.

[0067] In a feasible implementation manner, the step of performing Chinese character point-selection verification code recognition based on the Chinese character point-selection verification code network model to obtain a verification code recognition result described in this embodiment includes: inputting the Chinese character point-selection verification code into a deep convolutional network for feature extraction to obtain a deep feature, wherein the Faster-RCNN network comprises a deep convolutional network and a region candidate network, wherein the deep convolutional network comprises a convolutional layer, a ReLU activation layer, and a spatial region pyramid pooling layer; determining the region candidate coordinates corresponding to the Chinese character point-selection verification code through the region candidate network, wherein the region candidate network comprises candidate box classification and candidate box position correction; inputting the deep feature and the region candidate coordinates into the spatial region pyramid pooling layer to generate a region candidate deep feature; performing feature classification on the region candidate deep feature according to the convolutional layer and the ReLU activation layer to obtain a verification code recognition result of the Chinese character point-selection verification code.

[0068] It should be noted that the Deep Convolutional Network (DCN) is a deep learning network, which is mainly composed of multiple convolutional layers and other types of layers (such as pooling layers, fully connected layers, etc.). For example, the deep convolutional network of the Faster-RCNN network includes 13 convolutional layers, 13 ReLU activation layers and 4 spatial area pyramid pooling layers. The deep convolutional network is used to extract the feature map of the input image. The Faster-RCNN network classifies the feature map through the fully connected layer and outputs the verification code recognition result.

[0069] The convolution layer can extract the features of the Chinese character verification code through convolution operations, and can automatically learn different levels of feature representation, from low-level edge, texture and other features to high-level complex structural features.

[0070] It is understandable that the Region Proposal Network (RPN) can be a network for generating candidate regions that may contain targets in an image. RPN slides a window on the deep features of the Chinese character point verification code, and simultaneously predicts the probability of target existence corresponding to each window position and the region candidate coordinates and size of the candidate region. Among them, the candidate box classification part can realize whether all preset candidate boxes are Chinese characters through a convolution layer, two reshape layers and a softmax layer; the candidate box position correction part can calculate the offset of the candidate box through only one convolution layer to obtain accurate region candidate coordinates.

[0071] Among them, the Spatial Pyramid Pooling Layer (SPPL) can be a connection layer for solving the problem of non-fixed input image size. The Spatial Pyramid Pooling Layer divides the input depth features into regions of different sizes and proportions according to the region candidate coordinates, and then performs a pooling operation (such as maximum pooling or average pooling) on ​​each region, and combines the pooling results of these different regions to form a fixed-length region candidate depth feature. In this way, even if the size of the input image is different, a feature representation with a fixed dimension can be obtained after passing through the Spatial Pyramid Pooling Layer, thereby facilitating subsequent processing and classification tasks.

[0072] In this embodiment, first input the Chinese character point verification code, use the deep convolution network to extract the deep features of the input image, and obtain the deep features. Then generate the candidate box and the regional candidate coordinates of the Chinese character point verification code through the regional candidate network. For accurate classification, the extracted deep features and regional candidate coordinates can then be input into the spatial region pyramid pooling layer to generate regional candidate deep features. Finally, the overlapping convolutional layer and ReLU activation layer are used to realize the classification of regional candidate deep features, and the precise position and classification results of the Chinese character point verification code detection box are obtained. Through the Faster-RCNN network, the position and category of Chinese characters can be accurately identified, reducing the misrecognition rate.

[0073] In the technical solution provided in this embodiment, the YOLOv5 target detection algorithm and the Faster-RCNN network are used to accurately identify the position and category of Chinese characters, reduce the misrecognition rate, improve the overall recognition efficiency, and realize interference-resistant and highly accurate Chinese character point verification code recognition. The invention can be widely used in various network platforms that need to verify the identity of users, can improve the security level of verification codes, prevent attacks from automatic programs, and ensure the security of network services, thereby improving the reliability of data transaction Chinese character point verification user identity authentication and the security of data transactions.

[0074] Exemplarily, to facilitate understanding of the implementation process of the above-mentioned detection method for the security of Chinese character click verification codes based on deep learning, please refer to Figure 1 , Figure 1 which is the flowchart of the click verification code recognition based on the Faster-RCNN network of the present invention. The specific steps are as follows:

[0075] First, collect a number of Chinese character click verification code pictures and mark the positions of the Chinese characters therein to construct a Chinese character position detection data set.

[0076] Then, use the constructed Chinese character position detection data set to train a Chinese character position detection model based on the YOLOv5 object detection algorithm. The Chinese character position detection model is used to obtain the anchor boxes of each character.

[0077] Next, use the trained Chinese character position detection model to perform character detection on the click verification code pictures, save each Chinese character box as a separate picture and perform annotation, and collect the cropped Chinese character pictures and the corresponding annotation information to generate a Chinese character click verification code data set for training the Faster-RCNN network.

[0078] After that, use the constructed Chinese character click verification code data set to train the Faster-RCNN network to obtain a Chinese character click verification code network model.

[0079] Finally, the trained Chinese character click verification code network model can be used to perform click verification code recognition.

[0080] To facilitate understanding of the above click verification code recognition process, refer to Figure 2 , Figure 2 which is an example diagram of the click verification code provided by the present invention. As Figure 2 shown, there are four irregularly deformed Chinese characters "体", "验", "库", and "安" on the click verification code picture. After performing Chinese character detection using the Chinese character position detection model trained by the YOLOv5 object detection algorithm, separate Chinese character pictures of the four Chinese characters "安", "验", "库", and "体" can be obtained. Finally, through the Faster-RCNN network for character classification and recognition, the verification code recognition results of each Chinese character on the final click verification code picture can be obtained.

[0081] In the above verification code recognition link, the verification code recognition rate can be defined to evaluate the accuracy of model recognition through indicators such as precision and recall. Among them, the verification code recognition rate refers to the ratio of the correct recognition of the verification code by the automatic recognition system and human users respectively under specific test conditions. Specifically as follows:

[0082] 1. Define the verification code recognition rate.

[0083] First, the automatic recognition rate is defined: automatic recognition rate AR = total number of verification codes tested / number of verification codes correctly recognized × 100%. Then the human recognition rate is defined: human recognition rate HR = number of verification codes correctly recognized by humans / total number of verification codes tested × 100%.

[0084] 2. Set safety thresholds.

[0085] In order to ensure that the verification code is both secure and can be correctly identified by human users, the threshold standards are set as shown in Table 1:

[0086] Table 1

[0087] project Safety threshold Human recognition rate HR Greater than or equal to 90% Automatic recognition rate AR Less than 5%

[0088] For the above security threshold, in order to ensure the human user experience, the human recognition rate should be greater than or equal to 90%, that is, more than 90% of users can correctly recognize the verification code. In order to prevent automated attacks, the automatic recognition rate should be less than 5%, that is, only less than 5% of the verification codes can be recognized by the automated system.

[0089] 3. Verification code design and generation.

[0090] The design of the verification code needs to ensure its complexity while maintaining sufficient readability. Therefore, the following two design factors need to be considered: complexity design and randomness design.

[0091] For complexity design, assume that the generated verification code image contains N characters, where the changes of each character include random rotation, distortion, color change and background noise. The changes of each complexity are defined as follows:

[0092] Number of characters: The number of characters in the verification code is N = 4.

[0093] Character Distortion: The rotation angle range of each character.

[0094] Character color change: The RGB value of the color changes randomly, and the value range is RGB∈[0, 255].

[0095] Background interference: The background noise generation function fnoise(x,y) randomly generates noise points to interfere with the clarity of the background.

[0096] For random design, the verification code image generated each time has a high degree of randomness in the following aspects, such as the position, size and color of the characters are random, ensuring that each verification code generated is unique. At the same time, the background pattern and interference items (such as random lines, noise) can also change randomly.

[0097] 4. Testing and evaluation.

[0098] Through multiple rounds of testing, the verification code can be comprehensively evaluated to ensure its security.

[0099] The first step is to collect test samples. M = 10,000 verification code images of different styles can be generated, including various combinations of distortions and background interference. These verification codes are used for subsequent tests.

[0100] Then comes the testing process. For the human testing process, 50 users can be randomly selected and asked to identify 100 verification codes. The recognition results of each user are recorded as: HR i = Number of correct identifications / 100.

[0101] For the automatic recognition test process, the model trained by YOLOv5 and Faster-RCNN can be used to calculate its automatic recognition rate AR in 10,000 verification code images: AR = number of verification codes correctly recognized / 10,000 × 100%.

[0102] 5. Safety assessment criteria.

[0103] Based on the test results, the security of the verification code can be determined using the criteria shown in Table 2.

[0104] Table 2

[0105]

[0106]

[0107] As shown in Table 2, for high security, assuming that after testing, the human recognition rate HR = 92% and the automatic recognition rate AR = 4%, the verification code is judged to be high security and suitable for scenarios that require high protection. For medium security, assuming that the test results are human recognition rate HR = 85% and automatic recognition rate AR = 8%, the verification code is suitable for general use, but may need to be optimized in scenarios with higher security requirements. For low security, assuming that the test results are human recognition rate HR = 75% and automatic recognition rate AR = 20%, the verification code is judged to be low security and needs to be redesigned.

[0108] 6. Continuous optimization.

[0109] In order to ensure the long-term effectiveness and security of the verification code, regular monitoring and optimization can be performed. For the automatic recognition rate, the recognition rate of the automated recognition system is collected monthly, and its changes are recorded and analyzed. If the recognition rate increases, the complexity of the verification code may need to be adjusted. For the human recognition rate, a human user recognition test is conducted once a quarter to ensure that the human user experience is not affected.

[0110] Through the above method, the position and category of Chinese characters can be accurately identified, the misrecognition rate can be reduced, the overall recognition efficiency can be improved, and the high-accuracy Chinese character point verification code recognition with anti-interference can be realized. It can be applied to the scene of improving the security level of the verification code, thereby preventing the attack of automatic programs and ensuring the security of network services.

[0111] It should be noted that the above examples are only used to understand the present invention and do not constitute a limitation on the security detection method of the Chinese character point selection verification code based on deep learning of the present invention. More forms of simple transformations based on this technical concept are all within the protection scope of the present invention.

[0112] Based on the above method, this embodiment proposes a Chinese character point verification code security detection system based on deep learning, which executes and implements the Chinese character point verification code security detection method based on deep learning as described above.

[0113] In this embodiment, please refer to Figure 3 , Figure 3 The present invention is a structural diagram of a detection system for the security of Chinese character selection verification codes based on deep learning. The system includes: a client device 1, a distributed database device 2, a telescopic server 3, a cloud platform device, a detection system processing device 4, a real-time monitoring device 5 and a terminal visualization large screen 6.

[0114] The client device 1 is connected to the distributed database device 2, and the distributed database device 2 is connected to the telescopic server 3 (for example, server ESC), the cloud platform device, the detection system processing device 4, the real-time monitoring device 5 and the terminal visualization large screen 6 in turn.

[0115] In a feasible implementation manner, the client device 1 described in this embodiment is used to send a verification code request to the server and receive a verification code image, and to feed back the user's clicking behavior to the backend server.

[0116] It should be noted that the client device is a terminal device used for users to interact with the verification code. The client device is responsible for sending verification code requests to the server and receiving verification code images, as well as feeding back the user's selection behavior (i.e. the location of the selected Chinese characters) to the system. It is also responsible for sending the recognition results to the backend server for processing.

[0117] In a feasible implementation manner, the distributed database device 2 described in this embodiment is used to store user information, verification code data sets, model training data and verification code recognition results.

[0118] It should be noted that the distributed database device is used to store the data required by the system, including user information, verification code data sets, model training data, verification code recognition results, etc. The device supports the access of large-scale data through distributed storage technology, ensuring high data availability and high concurrent processing capabilities.

[0119] In a feasible implementation manner, the scalable server 3 in this embodiment is used to receive the verification code request from the client device and distribute the verification code request to the cloud platform device to dynamically adjust the allocation of computing resources.

[0120] It should be noted that the scalable server, as the computing engine of the system, is responsible for processing verification code requests from client devices. The scalable server supports dynamic expansion and can adjust the allocation of computing resources according to real-time load conditions. It is responsible for distributing verification code requests to cloud platform devices and exchanging data with other devices.

[0121] In a feasible implementation, the cloud platform device described in this embodiment is used to integrate the Chinese character position detection model constructed based on the YOLOv5 target detection algorithm and the Faster-RCNN network, and perform image processing and analysis of the verification code image, and feed back the verification code recognition result uploaded by the cloud platform device to the client device 1.

[0122] It should be noted that the cloud platform device processes and analyzes the verification code images transmitted to the cloud by integrating deep learning models (such as the Chinese character position detection model and Faster-RCNN network built based on the YOLOv5 target detection algorithm mentioned above). The cloud platform device not only supports the rapid recognition of verification code images, but also can perform training and reasoning tasks, and provide real-time feedback on recognition results.

[0123] In a feasible implementation manner, the detection system processing device 4 described in this embodiment is used to perform position detection and classification recognition on the verification code image through the Chinese character position detection model and the Faster-RCNN network, obtain the verification code recognition result, and upload it to the cloud platform device.

[0124] It should be noted that the detection system processing device is responsible for the real-time recognition and security detection of Chinese character verification codes. The device uses deep learning technology, such as the Chinese character position detection model built based on the YOLOv5 target detection algorithm to detect Chinese character positions, and uses the Faster-RCNN network to classify and recognize verification codes. The detection results will be uploaded to the cloud platform device to be returned to the telescopic server or client device.

[0125] In a feasible implementation manner, the real-time monitoring device 5 described in this embodiment is used to monitor the system status, and the system status includes the verification code recognition process, cloud platform resource usage and data transmission process; the terminal visualization large screen 6 is used to display the system operation status, verification code recognition results and administrator operation interface.

[0126] It should be noted that the real-time monitoring device is responsible for monitoring and managing the system status, including monitoring the verification code recognition process, cloud platform resource usage, and data transmission process; the device can give real-time alarms and automatically adjust system resource configuration to ensure the stability and efficient operation of the system. The terminal visualization screen is responsible for displaying the system operation status and test results. Through the visualization interface, the administrator can view the working status of each system module, the number of verification codes processed, the recognition accuracy and other information in real time, and perform system management and optimization.

[0127] The detection system for the security of Chinese character point verification codes based on deep learning provided by the present invention implements the detection method for the security of Chinese character point verification codes based on deep learning in the above-mentioned embodiment, and can solve the technical problem that the existing method is inefficient in recognizing Chinese character verification codes. Compared with the prior art, the beneficial effects of the detection system for the security of Chinese character point verification codes based on deep learning provided by the present invention are the same as the beneficial effects of the detection method for the security of Chinese character point verification codes based on deep learning provided by the above-mentioned embodiment, which will not be repeated here.

[0128] The above descriptions are only some embodiments of the present invention, and are not intended to limit the patent scope of the present invention. All equivalent structural changes made using the contents of the present invention's specification and drawings under the technical concept of the present invention, or directly / indirectly applied in other related technical fields, are included in the patent protection scope of the present invention.

Claims

1. A method for detecting the security of Chinese character point verification codes based on deep learning, characterized in that: The method comprises: Acquire Chinese character click verification code data, and mark the Chinese character positions in the Chinese character click verification code data to obtain a Chinese character position detection data set; Training a preset Chinese character position detection model according to the Chinese character position detection data set, wherein the Chinese character position detection model is constructed based on the YOLOv5 target detection algorithm; Perform Chinese character detection on the Chinese character click verification code image through the trained Chinese character position detection model to obtain a Chinese character click verification code dataset; A preset Faster-RCNN network is trained according to the Chinese character click verification code data set to obtain a Chinese character click verification code network model; Based on the Chinese character selection verification code network model, Chinese character selection verification code recognition is performed to obtain a verification code recognition result.

2. The method according to claim 1, characterized in that The step of performing Chinese character detection on the Chinese character point verification code picture by using the trained Chinese character position detection model to obtain a Chinese character point verification code data set includes: Perform Chinese character detection on the Chinese character point verification code picture by using the trained Chinese character position detection model to obtain the Chinese character position and position coordinates in the Chinese character point verification code picture; According to the position coordinates, a Chinese character rectangular frame corresponding to each Chinese character position is cut out from the Chinese character point selection verification code picture, and the Chinese character rectangular frame is converted into a Chinese character picture; Mark the Chinese characters corresponding to the Chinese character images and generate corresponding txt files; The Chinese character images and the txt files are collected, and the file names of the Chinese character images are matched with the file names of the txt files to generate a Chinese character point verification code data set.

3. The method according to claim 2, characterized in that The step of performing Chinese character click verification code recognition based on the Chinese character click verification code network model to obtain a verification code recognition result comprises: Inputting the Chinese character point verification code into a deep convolutional network for feature extraction to obtain deep features, wherein the Faster-RCNN network comprises a deep convolutional network and a region candidate network, and the deep convolutional network comprises a convolutional layer, a ReLU activation layer, and a spatial region pyramid pooling layer; Determine the region candidate coordinates corresponding to the Chinese character point verification code through the region candidate network, wherein the region candidate network is composed of candidate box classification and candidate box position correction; Inputting the depth feature and the region candidate coordinates into the spatial region pyramid pooling layer to generate region candidate depth features; Feature classification is performed on the candidate region depth features according to the convolution layer and the ReLU activation layer to obtain a verification code recognition result of the Chinese character point selection verification code.

4. The method according to claim 3, characterized in that The step of training a preset Faster-RCNN network according to the Chinese character click verification code data set to obtain a Chinese character click verification code network model includes: Resampling a plurality of Chinese character images in the Chinese character click verification code data set to obtain resampled images of the same size as the Chinese character images; The resampled image and the txt file in the Chinese character click verification code data set are input into the preset Faster-RCNN network for training to obtain the Chinese character click verification code network model.

5. The method according to claim 4, characterized in that The deep convolutional network of the Faster-RCNN network includes 13 convolutional layers, 13 ReLU activation layers and 4 spatial area pyramid pooling layers. The deep convolutional network is used to extract the feature map of the input image. The Faster-RCNN network classifies the feature map through a fully connected layer and outputs a verification code recognition result.

6. A Chinese character point verification code security detection system based on deep learning, characterized in that: The detection system for the security of Chinese character point verification code based on deep learning implements the detection method for the security of Chinese character point verification code based on deep learning as claimed in claim 1, and the system includes: client device, distributed database device, telescopic server, cloud platform device, detection system processing device, real-time monitoring device and terminal visualization large screen; The client device is connected to the distributed database device, and the distributed database device is connected to the telescopic server, the cloud platform device, the detection system processing device, the real-time monitoring device and the terminal visualization large screen in sequence.

7. The system according to claim 6, characterized in that The client device is used to send a verification code request to the server and receive a verification code image, and to feed back the user's clicking behavior to the backend server; The distributed database device is used to store user information, verification code data sets, model training data and verification code recognition results; The retractable server is used to receive the verification code request from the client device and distribute the verification code request to the cloud platform device to dynamically adjust the allocation of computing resources; The cloud platform device is used to integrate the Chinese character position detection model built based on the YOLOv5 target detection algorithm and the Faster-RCNN network, perform image processing and analysis of the verification code image, and feed back the verification code recognition result uploaded by the cloud platform device to the client device; The detection system processing device is used to perform position detection and classification recognition on the verification code image through the Chinese character position detection model and the Faster-RCNN network, obtain the verification code recognition result, and upload it to the cloud platform device; The real-time monitoring device is used to monitor the system status, which includes the verification code recognition process, cloud platform resource usage and data transmission process; The terminal visualization large screen is used to display the system operation status, verification code recognition results and administrator operation interface.

Images