Target false alarm detection method, electronic device and computer readable storage medium

Abstract

This application discloses a target false alarm detection method, an electronic device, and a computer-readable storage medium. The method includes: obtaining the target location and corresponding target category of a target in an image to be detected, whereby the target category has a corresponding category value; in response to a target category value being lower than a first preset threshold, extracting a false alarm detection image corresponding to the target location from the image to be detected using the target location; and inputting the false alarm detection image into a false alarm filtering network for category detection to correct the target category. Through the above method, this application solves the technical problem of numerous false alarms in target detection.

Description

Technical Field

[0001] This application relates to the field of computer vision, and in particular to a method for detecting false alarms of targets, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Object detection is increasingly widely used in various fields. It utilizes computer vision, pattern recognition, and digital image processing technologies to automatically process image data, locate, identify, and track targets within the image, then analyze and determine the category of the detected targets before performing corresponding processing operations. However, object detection typically relies on large amounts of labeled data, which often requires lengthy training times, limiting the development of object detection applications. Therefore, few-shot object detection has emerged. It learns from a small number of samples to locate and identify targets in the image to be detected. There are two implementation methods for few-shot object detection: one based on fine-tuning and the other based on template registration. While few-shot object detection can solve the problem of time-consuming and laborious data labeling, it tends to generate more false positives due to the limited new class information it learns. Summary of the Invention

[0003] The main purpose of this application is to provide a target false alarm detection method, electronic device, and computer-readable storage medium that can solve the technical problem of high false alarm rates in target detection.

[0004] To address the aforementioned technical problems, the first technical solution adopted in this application is: providing a target false alarm detection method. This method includes: obtaining the target location and corresponding target category of a target in the image to be detected, whereby the target category has a corresponding category value; responding to a target category value being lower than a first preset threshold, extracting a false alarm detection image corresponding to the target location from the image to be detected using the target location; and inputting the false alarm detection image into a false alarm filtering network for category detection to correct the target category.

[0005] To address the aforementioned technical problems, the second technical solution adopted in this application is to provide an electronic device. This electronic device includes a memory and a processor. The memory stores program data, which can be executed by the processor to implement the method described in the first technical solution.

[0006] To address the aforementioned technical problems, the third technical solution adopted in this application is to provide a computer-readable storage medium. This computer-readable storage medium stores program data and can be executed by a processor to implement the method described in the first technical solution.

[0007] The beneficial effects of this application are as follows: By inputting the image to be detected into the target detection network for detection, the location and corresponding category of the target in the image can be obtained. Each target category output has a corresponding category value, representing the probability that the target category is correctly identified. When the output target category value is lower than a first preset threshold, it indicates that there is a possibility of misclassification of the target category. Therefore, using the obtained target location, a false positive detection image is extracted from the image to be detected at the corresponding location. This false positive detection image is then input into the false positive filtering network for re-classification, thereby correcting the target category. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the first embodiment of the target false alarm detection method of this application;

[0010] Figure 2 This is a flowchart illustrating the second embodiment of the target false alarm detection method of this application;

[0011] Figure 3 This is a flowchart illustrating the third embodiment of the target false alarm detection method of this application;

[0012] Figure 4 This is a schematic diagram of the training process of the false alarm detection method for the target of this application;

[0013] Figure 5 This is a schematic diagram of the detection process of the false alarm detection method of this application;

[0014] Figure 6 This is a schematic diagram of the structure of the first embodiment of the electronic device of this application;

[0015] Figure 7 This is a schematic diagram of the structure of the first embodiment of the computer-readable storage medium of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0017] The terms "first," "second," etc., used in this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0018] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0019] Reference Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the false alarm detection method for the target of this application. It includes the following steps:

[0020] S11: Obtain the target location and corresponding target category in the image to be detected based on the image to be detected.

[0021] The image to be detected is first input into the object detection network to obtain the initial category recognition result. This initial result includes the target location and its corresponding category. For example, for the image to be detected, the number 1 can be detected. The location of 1 in the image and its corresponding category A can be obtained. Each category corresponds to a category value, which can be a number between 0 and 1, with a minimum of 0 and a maximum of 1, such as 0.7, 0.8, etc., representing the probability that the target category is correctly identified. Typically, for a detected target, at least one category can be identified, and the category values ​​of all these categories are summed to 1. The category corresponding to the largest category value is then used as the output category for that detected target. For example, when detecting 1, if the category value for classifying it as A is 0.7 and the category value for classifying it as B is 0.3, then A is output as the target category for 1.

[0022] S12: In response to the target category value being lower than the first preset threshold, a false alarm detection image corresponding to the target location is extracted from the image to be detected using the target location.

[0023] If the category value of the detected target is lower than a first preset threshold, it is determined that there may be a false alarm in the classification of the target category, and the target needs to be re-identified and classified. Based on the target location corresponding to the detected target obtained from the first category identification result, an image of the region corresponding to the target location is obtained from the image to be detected as the false alarm detection image of the target, which is used for the second category identification.

[0024] In this embodiment, the first preset threshold can generally be set to a value in the range of 0.5-1.

[0025] The first preset threshold can be set based on the actual training performance of the object detection network. For example, when training the object detection network to recognize category A, if the network achieves the highest recognition accuracy for training images when the first preset threshold is 0.7, then this value can be used as the first preset threshold for the recognition of that category.

[0026] S13: Input the false alarm detection image into the false alarm filtering network for category detection to correct the target category.

[0027] The acquired false alarm detection images are input into a false alarm filtering network for a second category recognition. The target category from the first recognition is then corrected based on the recognition results. If the first target category is correctly identified, it is used as the final result; otherwise, the second target category is used as the final result.

[0028] In this embodiment, the image to be detected is input into a target detection network for detection, which obtains the location and corresponding category of the target in the image. Each target category outputs a corresponding category value, representing the probability that the target category is correctly identified. When the output target category value is lower than a first preset threshold, it indicates a possibility of misclassification. Therefore, using the obtained target location, a false positive detection image is extracted from the image to be detected at the corresponding location. This false positive detection image is then input into a false positive filtering network for re-classification, thereby correcting the target category.

[0029] In one embodiment, the image to be detected is preprocessed before being input into the target detection network. Preprocessing may include three parts: grayscale conversion, geometric transformation, and image enhancement. Grayscale conversion methods may include component methods, maximum value methods, average value methods, and weighted average methods. Geometric transformations may include a set of operations such as translation, transpose, mirroring, rotation, and scaling. Image enhancement may include frequency domain methods and spatial domain methods. Preprocessing can eliminate irrelevant information in the image, restore useful real information, enhance the detectability of relevant information, and simplify data to the maximum extent, thereby improving the reliability of feature extraction and matching recognition.

[0030] In one embodiment, in response to the target category value being no lower than a first preset threshold, no correction is made to the target category. When the output target category value is no lower than the first preset threshold, it indicates that the judgment of the target category is not considered to be misjudged, and the first category recognition result is directly used as the final result.

[0031] In one embodiment, each target category has its corresponding first preset threshold. During training, training can be performed for the corresponding categories to obtain multiple first preset thresholds for each category. Then, during the false alarm filtering process of acquiring the image to be detected, in response to a target category value being lower than the first preset threshold corresponding to the target category, a false alarm detection image corresponding to the target location is extracted from the image to be detected using the target location. The obtained false alarm detection image is then re-inputted into the false alarm filtering network for category detection.

[0032] In one embodiment, inputting a false alarm detection image into a false alarm filtering network for category detection to correct the target category includes: correcting the target category with the corrected category in response to the category value of the corrected category output by the false alarm filtering network being higher than a second preset threshold; and determining that the category recognition of the image to be detected is incorrect in response to the category value of the corrected category output by the false alarm filtering network being lower than the second preset threshold.

[0033] After the false alarm detection image is input into the false alarm filtering network, the network also outputs the corrected category and corresponding category value based on the false alarm detection image. The second preset threshold can be set in the range of 0.5-1, and can be a value less than the first preset threshold. The second preset threshold can be set to 0.5. When the category value of the corrected category output by the false alarm filtering network is higher than the second preset threshold, the corrected category output by the false alarm filtering network is used for judgment. If it is the same as the target category detected in the first detection, the first judgment is correct, and the target category is output. If it is different from the target category detected in the first detection, the first judgment is incorrect, and the corrected category is output. When the category value of the corrected category output by the false alarm filtering network is lower than the second preset threshold, it indicates that the image to be detected contains too much new information, and neither the target detection network nor the false alarm detection network can correctly identify and classify it, resulting in an incorrect category identification for the image to be detected.

[0034] Reference Figure 2 , Figure 2 This is a flowchart illustrating a second embodiment of the false alarm detection method for the target of this application. The method is a further extension of step S11, and includes the following steps:

[0035] S21: Input the image to be detected into the backbone network of the target detection network to obtain the feature map of the image to be detected.

[0036] S22: Input the feature map into the RPN network in the object detection network to obtain the location of the target to be determined.

[0037] S23: Pool the features obtained based on the feature map and the target location to obtain regional features of the same size.

[0038] S24: Input the region features into the feature extractor to obtain aggregated features.

[0039] S25: Obtain the target location and determine the target category based on aggregated features.

[0040] The image to be detected is input into the object detection network to obtain the target location and target category. First, the image is input into the backbone network to obtain a feature map. The backbone network is a feature extraction network used to extract image features. After obtaining the feature map, the RPN network is used to obtain the location of the target. The location of the target can be a set of detection boxes. Since the size of the detection boxes may be inconsistent, after obtaining the features based on the detection boxes and feature map, the features are pooled to make them uniform in size. Then, all the features are aggregated, and the aggregated features are used for feature classification and regression. Feature classification is used to obtain the target classification result, and regression is used to adjust the location of the target to be detected, making it closer to the true target detection box, thus obtaining the target location.

[0041] In one embodiment, the false alarm filtering network is trained by acquiring the location of the target to be determined and the category information corresponding to the target location from the output of the RPN network.

[0042] Reference Figure 3 , Figure 3 This is a flowchart illustrating a third embodiment of the false positive detection method of this application. The method extends the training steps of the false positive filtering network and includes the following steps:

[0043] S31: Obtain the location of the target to be determined and the category information corresponding to the target location obtained by the RPN network based on the training image.

[0044] During the training of the object detection network, the locations of the target to be determined are obtained when positive and negative samples pass through the RPN network, and the categories of the positive and negative samples are used as the category information corresponding to their target locations.

[0045] S32: Obtain the training detection image based on the location of the target to be determined and the training image.

[0046] Extract the training detection image from the training image based on the location of the target to be detected. The location of the target to be detected can be a detection box.

[0047] S33: Input the training detection images and corresponding category information into the false alarm filtering network for category detection training.

[0048] The location of the target to be determined has its corresponding category information, and the training detection image corresponds to the location of the target; therefore, the training detection image will also have its corresponding category information. The obtained training detection image and its corresponding category information are input into the false alarm filtering network to train the false alarm filtering network.

[0049] The false positive filtering network uses the positive and negative sample classes of the RPN as input to the false positive filtering module, eliminating the need for long-term collection of false positive data. Furthermore, negative sample classes such as background can be learned by the false positive filtering network on both base and new classes, resulting in better generalization performance compared to deep cosine metric models.

[0050] In one embodiment, the false positive filtering network is trained using the BYOL unsupervised method. The network includes convolutional layers and fully connected layers. The convolutional layers are used for feature extraction, and the fully connected layers are used for category recognition.

[0051] The following specific embodiment will be used to illustrate the technical solution of this application in more detail.

[0052] like Figure 4 As shown, Figure 4 This is a schematic diagram of the training process for the false alarm detection method of this application.

[0053] First, the input image is preprocessed. Then, the processed image is fed into the object detection network, and object detection training is performed on the base class.

[0054] Then, detection boxes for positive and negative sample classes are obtained from the RPN network. False alarm detection images are obtained from the image to be detected based on the detection boxes. Finally, the corresponding false alarm detection images are trained based on the category information of positive and negative sample classes.

[0055] When fine-tuning with a new class, repeat the above steps.

[0056] like Figure 5 As shown, Figure 5 This is a schematic diagram of the detection process of the target false alarm detection method of this application. During detection, the target bounding box position (obtained based on the RPN network and regression boxes) and target category are obtained through the target detection network. When the category value of the target category is lower than a first preset threshold, the corresponding region is found on the image to be detected based on the target bounding box position and cropped to obtain a false alarm detection image. The false alarm detection image is then classified and identified to correct the target category. When the category value of the target category is higher than the first preset threshold, no processing is performed, and the target category is used as the output category of the detection image.

[0057] like Figure 6 As shown, Figure 6 This is a schematic diagram of the structure of the first embodiment of the electronic device of this application.

[0058] The electronic device includes a processor 110 and a memory 120.

[0059] Processor 110 controls the operation of electronic devices. Processor 110 may also be referred to as a CPU (Central Processing Unit). Processor 110 may be an integrated circuit chip with signal sequence processing capabilities. Processor 110 may also be a general-purpose processor, a digital signal sequence processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0060] The memory 120 stores the instructions and program data required for the processor 110 to operate.

[0061] The processor 110 is used to execute instructions to implement the methods provided in any embodiment and possible combination of the target false alarm detection methods described above in this application.

[0062] like Figure 7 As shown, Figure 7This is a schematic diagram of the structure of the first embodiment of the computer-readable storage medium of this application.

[0063] One embodiment of the readable storage medium of this application includes a memory 210 that stores program data that, when executed, implements the method provided in any embodiment and possible combination of the target false alarm detection method of this application.

[0064] The memory 210 may include a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or other media that can store program instructions. Alternatively, it may be a server that stores the program instructions, which can send the stored program instructions to other devices for execution or execute the stored program instructions itself.

[0065] In summary, by inputting the image to be detected into the target detection network, the location and corresponding category of the target in the image can be obtained. Each target category outputs a corresponding category value, representing the probability that the target category is correctly identified. When the output target category value is lower than a first preset threshold, it indicates a possibility of misclassification. Therefore, using the obtained target location, a false positive detection image is extracted from the image to be detected at the corresponding location. This false positive detection image is then input into the false positive filtering network for re-classification, thereby correcting the target category.

[0066] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0067] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0068] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0069] If the integrated units in the other embodiments described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0070] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for detecting false alarms of a target, characterized in that, The method, applied to small sample target detection, includes: Based on the image to be detected, the target location and corresponding target category of the target in the image to be detected are obtained, and the target category has a corresponding category value; In response to the target category value being lower than a first preset threshold corresponding to the target category, a false alarm detection image corresponding to the target location is extracted from the image to be detected using the target location; wherein, each target category has its corresponding first preset threshold; The false alarm detection image is input into a false alarm filtering network for category detection to correct the target category; The training of the false positive filtering network includes: Implementing target detection training on a base class includes: obtaining the location of the target to be determined, the positive and negative sample classes obtained by the RPN network in the target detection network based on the training image, and the category information corresponding to the location of the target to be determined; A training detection image is obtained based on the undetermined target location and the training image; The training detection images and the corresponding category information are input into the false alarm filtering network for category detection training. When making adjustments using the new class, repeat the above steps.

2. The method according to claim 1, characterized in that, The step of obtaining the target location and corresponding target category of the target in the image to be detected based on the image to be detected includes: The image to be detected is input into the backbone network of the target detection network to obtain the feature map of the image to be detected; The feature map is input into the RPN network in the target detection network to obtain the location of the target to be determined. Pooling is performed on the features obtained based on the feature map and the target location to obtain region features of the same size; The region features are input into a feature extractor to obtain aggregated features; The target location and target category are determined based on the aggregated features.

3. The method according to claim 1, characterized in that, The false alarm filtering network is trained using the BYOL unsupervised method.

4. The method according to claim 1, characterized in that, The step of inputting the false alarm detection image into a false alarm filtering network for category detection to correct the target category includes: In response to the false alarm filtering network outputting a correction category value that is higher than a second preset threshold, the target category is corrected using the correction category; If the category value of the corrected category output by the false alarm filtering network is lower than the second preset threshold, it is determined that the category recognition of the image to be detected is incorrect.

5. The method according to claim 2, characterized in that, Before inputting the image to be detected into the backbone network of the target detection network to obtain the feature map of the image to be detected, the following steps are included: The detected image is preprocessed.

6. An electronic device, characterized in that, It includes a memory and a processor, the memory being used to store program data, the program data being executable by the processor to implement the method as described in any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, It stores program data and can be executed by a processor to implement the method as described in any one of claims 1-5.

Images