A target detection method, system, device and medium based on brain-like feedback

Abstract

The application discloses a target detection method, system, device and medium based on brain-like feedback, and the method comprises the following steps: establishing a brain-like feedback network based on an IECA module and an APC module; performing environment severity evaluation processing on an input image by using the IECA module of the brain-like feedback network to obtain an evaluation result; generating a first additional feature for a target detection process based on the evaluation result; selecting an interest region in the input image by using the APC module of the brain-like feedback network and performing non-modal prediction completion processing to obtain a second additional feature for the target detection process; and inputting the input image, the first additional feature and the second additional feature into a target detection model for target detection; the application can simulate the human brain visual feedback mechanism in the target detection task in a harsh environment based on the brain-like feedback network, and achieve a more accurate target detection effect.

Description

Technical Field

[0001] This invention relates to the field of target detection technology, specifically to the field of autonomous driving, and particularly to a target detection method, system, device, and medium based on brain-like feedback. Background Technology

[0002] Existing object detection methods typically improve the model's feature extraction ability by reducing the impact of severe weather on images, i.e., by improving image quality to enhance detection accuracy. However, this approach actually contradicts the cognitive mechanisms of human vision, resulting in low detection realism and robustness. Severe weather conditions, with their low contrast and varying lighting, significantly blur object features, causing general models trained on high-quality data to often fail to achieve good detection results under such conditions.

[0003] Therefore, existing target detection methods have the following problems when dealing with detection tasks under harsh environmental conditions:

[0004] Firstly, existing methods contradict human visual cognitive processes, resulting in poor detection realism and robustness, which to some extent affects the accuracy of detection.

[0005] Secondly, existing methods have many algorithm parameters and complex settings, making them unable to adapt to different weather conditions, resulting in poor universality and low convenience. Summary of the Invention

[0006] The purpose of this invention is to address the aforementioned problems in the prior art by providing a target detection method, system, device, and medium based on brain-like feedback, thereby resolving the issue that existing target detection methods in adverse weather conditions contradict human visual cognitive processes, resulting in poor detection accuracy, realism, robustness, and applicability.

[0007] To solve the above-mentioned technical problems, the specific technical solution of the present invention is as follows:

[0008] On the one hand, the present invention provides a target detection method based on brain-like feedback, comprising the following steps:

[0009] Feedback network construction: Establish a brain-like feedback network based on the IECA module and the APC module;

[0010] Neuromorphic feedback processing: When using object detection models for environmental recognition in adverse weather conditions:

[0011] The IECA module of the aforementioned neuromorphic feedback network is used to assess the environmental severity of the input image, and an assessment result is obtained; based on the assessment result, a first additional feature is generated for the target detection process.

[0012] The APC module of the brain-like feedback network selects a region of interest in the input image and performs non-modal prediction completion processing to obtain a second additional feature for the target detection process.

[0013] Feature connection: The input image, the first additional feature, and the second additional feature are input into the target detection model for target detection.

[0014] As an improved approach, the IECA module uses ResNet as a feature extractor;

[0015] The last layer of the IECA module has a linear layer with an output dimension of 1;

[0016] The loss function of the IECA model is L1 loss.

[0017] As an improved approach, before the IECA module employing the neuromorphic feedback network performs environmental severity assessment on the input image, the following steps are included:

[0018] Obtain the dataset;

[0019] Noise is added to the dataset to obtain a simulated foggy weather dataset;

[0020] The IECA module was trained using the simulated fog weather dataset.

[0021] As an improved approach, the APC module employing the neuromorphic feedback network selects a region of interest in the input image and performs non-modal prediction completion processing to obtain a second additional feature for the target detection process, including:

[0022] The input image is input to the APC module;

[0023] The backbone network of the APC module outputs a one-hot vector.

[0024] Select the one-hot vector with the highest score as the activation class;

[0025] The weight value of each pixel position in the image region of the activated class is calculated using the weight calculation formula;

[0026] Calculate the auxiliary coefficients based on the weight values ​​and coefficient calculation formulas;

[0027] Calculate the positioning map based on the aforementioned auxiliary coefficients and positioning map calculation formula;

[0028] The obtained positioning image is retained as a grayscale image, and the size of the grayscale image is adjusted to fit the size of the input image;

[0029] The adjusted grayscale image is used as the second additional feature.

[0030] As an improved solution, the weight calculation formula is:

[0031]

[0032] in:

[0033]

[0034]

[0035] The weight value of the activation class in the k-th layer at position (i,j);

[0036] i∈[0,w], and j∈[0,w];

[0037] Let be the p-th order partial derivative of the class fraction in the k-th layer with respect to the position (i,j) of class c;

[0038] n is the number of samples used for gradient estimation.

[0039] As an improved solution, the formula for calculating the coefficient is:

[0040]

[0041] in:

[0042] The auxiliary coefficient;

[0043] A k (i,j) represents the activation value of node k at position (i,j) in the target layer;

[0044] Y(c) is the output score of the model for class c before softmax;

[0045]

[0046] The formula for calculating the location map is:

[0047]

[0048] Where: L (c) (x,y) is the positioning map.

[0049] As an improved approach, the step of inputting the input image, the first additional feature, and the second additional feature into the target detection model for target detection includes:

[0050] The first additional feature is connected to one dimension of the input image, making the size of the input image h*w*(d+1), for the purpose of target detection;

[0051] The second additional feature is used as a mask to cover the channel dimension of the input image, so that the shape of the input image becomes h*w*(d+1+1), which is used for target detection;

[0052] Where w is the width of the input image, h is the height of the input image, and d is the number of channels in the input image.

[0053] On the other hand, the present invention also provides a target detection system based on brain-like feedback, comprising:

[0054] The module includes a feedback network construction module, a brain-like feedback processing module, and a feature connection module.

[0055] The feedback network construction module is used to establish a brain-like feedback network based on the IECA module and the APC module;

[0056] The neuromorphic feedback processing module is used to: evaluate the environmental severity of the input image using the IECA module of the neuromorphic feedback network when performing environmental recognition under severe weather conditions using a target detection model; generate a first additional feature for the target detection process based on the evaluation result; and select a region of interest in the input image using the APC module of the neuromorphic feedback network and perform nonmodal prediction completion processing to obtain a second additional feature for the target detection process.

[0057] The feature connection module is used to input the input image, the first additional feature, and the second additional feature into the target detection model for target detection.

[0058] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the target detection method based on neuromorphic feedback.

[0059] On the other hand, the present invention also provides a computer device, the computer device including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; wherein:

[0060] The memory is used to store computer programs;

[0061] The processor is configured to execute the steps of the neuromorphic feedback-based target detection method by running a program stored in the memory.

[0062] The beneficial effects of the technical solution of this invention are:

[0063] 1. The target detection method based on brain-like feedback described in this invention can realize the design based on brain-like feedback network. In target detection tasks under harsh environments, it simulates the human brain's visual feedback mechanism, performs predictive encoding processing and non-modal perception completion processing on targets under adverse environmental conditions, achieves more accurate target detection results, improves detection authenticity and robustness, and has high application value.

[0064] 2. The target detection system based on brain-like feedback described in this invention can achieve the design based on brain-like feedback network through the cooperation of feedback network construction module, brain-like feedback processing module and feature connection module. In target detection tasks under harsh environments, it simulates the human brain's visual feedback mechanism to predict and encode targets under adverse environmental conditions and perform non-modal perception completion, thereby achieving more accurate target detection results, improving detection authenticity and robustness, and has high application value.

[0065] 3. The computer-readable storage medium of the present invention can enable the guided feedback network construction module, the neuromorphic feedback processing module and the feature connection module to cooperate, thereby realizing the target detection method based on neuromorphic feedback of the present invention. Furthermore, the computer-readable storage medium of the present invention also effectively improves the operability of the target detection method based on neuromorphic feedback.

[0066] 4. The computer device described in this invention can store and execute the computer-readable storage medium, thereby realizing the target detection method based on neuromorphic feedback described in this invention. Attached Figure Description

[0067] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0068] Figure 1 This is a flowchart illustrating the target detection method based on neuromorphic feedback as described in Embodiment 1 of the present invention;

[0069] Figure 2 This is a detailed flowchart of the target detection method based on neuromorphic feedback as described in Embodiment 1 of the present invention;

[0070] Figure 3 This is a schematic diagram of the logical architecture of BFNet in the target detection method based on brain-like feedback described in Embodiment 1 of the present invention;

[0071] Figure 4 This is a schematic diagram of the architecture of the target detection system based on brain-like feedback as described in Embodiment 2 of the present invention;

[0072] Figure 5 This is a schematic diagram of the structure of the computer device described in Embodiment 4 of the present invention;

[0073] The markings in the attached diagram are explained as follows:

[0074] 1501. Processor; 1502. Communication interface; 1503. Memory; 1504. Communication bus. Detailed Implementation

[0075] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0076] In the description of this invention, it should be noted that the embodiments described in this invention are only some embodiments of this invention, not all embodiments; based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0077] The terms "first," "second," etc., used in this specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0078] Example 1

[0079] This embodiment provides a target detection method based on brain-like feedback, such as Figures 1-3 As shown, it includes the following steps:

[0080] As one embodiment of the present invention, this method introduces the IECA (Initial Environmental Cognition Assessment) module and the APC (Amodal Predictive Completion) module into the YOLO model (i.e., the object detection model), thereby simulating the human brain's realistic visual feedback mechanism for object detection under adverse weather conditions. In this method, the IECA module and the APC module constitute BFNet (Brain-inspired Feedback). This method utilizes a brain-like feedback network (BNRF) to achieve more realistic and accurate object detection. In real-world scenarios, when humans encounter adverse or severe environmental conditions, they typically first assess the severity of the current conditions, such as fog density, to obtain global information. Then, humans select regions of interest based on their own interests and use non-modal perception for object detection and classification. Therefore, this method first passes the input image through an IECA module to assess the global severity of the environment. The severity is then filled into a matrix matching the image size and used as an additional channel in the image. Subsequently, the image passes through an APC module to simulate non-modal perception, generating a grayscale image of the Region of Interest (ROI). This grayscale image is also added to the image as an additional channel. By connecting these two feature maps along the channel dimension, the method simulates the brain's predictive coding mechanism, applying prior knowledge to process the fuzzy information obtained by the visual system. Finally, the processed image is input into the object detection model to obtain the object detection results.

[0081] As one embodiment of the present invention, the aforementioned non-modal perception refers to the following: Under harsh or adverse environmental conditions, the human body's ability to distinguish objects through visual perception is limited, leading to increased ambiguity and uncertainty. To accurately understand / judge the environment under such conditions, the brain employs a non-modal perception mechanism to predict the overall shape of objects. This mechanism allows the brain to infer the approximate shape of occluded objects, thus providing a more comprehensive understanding. Similarly, the human visual system has the ability to perceive the complete physical structure of objects, accurately predicting their shape even when they are partially or severely occluded. Therefore, in real-world situations, when faced with ambiguous visual input, the aforementioned mechanism enables the brain to actively explore using its inherent hierarchical knowledge of objects to provide multiple possible interpretations. This hierarchical structure is based on previously acquired experiential knowledge and cognitive models, continuously adjusting and optimizing the accuracy of the final interpretation through a feedback mechanism. In the context of ambiguous information, the brain enhances its understanding of complex visual input by integrating prior knowledge, filling in missing information, and constructing speculative interpretations. Based on these processes, the brain enhances its comprehension ability when facing ambiguous information, helping the human body to identify objects in harsh or adverse environments.

[0082] As one embodiment of the present invention, based on the above principles, this method includes the following steps:

[0083] S100. Introduce BFNet based on IECA and APC modules into the target detection model;

[0084] S200. When dealing with environmental recognition under severe weather conditions, the IECA module is used to evaluate the severity of the environment in the original environmental image to obtain the corresponding evaluation result. Based on this evaluation result, the first additional feature for final target detection is generated.

[0085] S300: The APC module is used to select the corresponding region of interest in the original environment image and perform non-modal prediction completion processing to obtain the second additional feature for final target detection;

[0086] S400: Input the original environmental image, the first additional feature, and the second additional feature into the target detection model for subsequent target detection to improve the accuracy and realism of the detection.

[0087] In one embodiment of the present invention, step S200 specifically includes the following operations:

[0088] Specifically, the IECA module uses ResNet as the feature extractor, and the last layer of the IECA module has a linear layer, producing an output dimension of 1. Furthermore, the loss function of the IECA model is L1 loss because the evaluation of global severity does not belong to the classification task, but to the data regression task.

[0089] After evaluation by the IECA module, the corresponding severity value is obtained. In this method, the severity value is filled into a matrix of the same size as the input bin, which serves as global information for subsequent object detection, simulating human feedback and judgment of global information; this information is the aforementioned first additional feature.

[0090] In one embodiment of the present invention, step S200 further includes:

[0091] To accurately evaluate raw environmental images using IECA, it can be trained on a corresponding dataset before use. Since quantifying fog density under real adverse conditions usually requires additional environmental indicators, which are difficult to obtain, this method uses noise added to a normal dataset to simulate fog weather data. By introducing noise, the degradation level required to effectively introduce visual information is achieved, enabling quantitative measurement of adverse environments and simulating the brain's evaluation response to adverse conditions.

[0092] Specifically, the method for adding noise is as follows:

[0093] For the images in the original dataset, defined as x∈R h*w*d The following formula is used to calculate the addition of fog noise:

[0094] I(x)=J(x)e -β d(x)+A(1-e -β ); formula (1);

[0095] In formula (1):

[0096] J(x) represents the original scene pixels;

[0097] A is set to 0.5;

[0098] β = 0.01·i + 0.05;

[0099] i is a positive integer, ranging from 0 to 9;

[0100] In formula (1): Where ρ represents the L2 distance from the current pixel to the center pixel, and the beta value of each input image is used as a parameter for evaluation by the IECA module;

[0101] In one embodiment of the present invention, step S300 further includes:

[0102] When performing APC analysis, the APC backbone network outputs a one-hot vector. Where C represents the number of classes in the dataset; to visualize the region of interest (ROI) of each input image, the output of the aforementioned backbone network is obtained. Then, the class with the highest score is selected as the activation class c; where, In this method, the ArgMax function is used to select the class with the highest probability as the activation class; then, the weight of each pixel position in the image region of the activation class is calculated using formula (2), and auxiliary coefficients are calculated based on the weights. Then, the localization map is calculated based on the auxiliary coefficients; the final calculated localization map is kept as a grayscale image, and the size of the grayscale image is adjusted to fit the size of the original image x of the input target detection model; the grayscale image after adjustment is used as the second additional feature mentioned above; in this step, the range and approximate shape of the object of interest in the scene are estimated by the APC module, and this information is processed together with the input image x for downstream tasks, so that the modified input image can be used in the subsequent target detection process;

[0103] In one embodiment of the present invention, step S300 further includes: for the APC module, pre-training is performed using a dataset; wherein M3TR is selected as the backbone for training the multi-label classification task; in the CAM module (i.e., the class activation mapping module), Smooth Grad-CAM is used as the visualization method. Based on this method, a smooth gradient-based class activation map is provided, which is then used to highlight the discriminative regions corresponding to the selected classes.

[0104] In one embodiment of the present invention, step S300 further includes:

[0105] When calculating the weight of each pixel position, formula (2) is as follows:

[0106]

[0107] in,

[0108] in,

[0109] in, Let be the weight of class c in the k-th layer at position (i,j);

[0110] Where i∈[0,w] and j∈[0,w];

[0111] in, Let be the p-th order partial derivative of the class fraction in the k-th layer with respect to the position (i,j) of class c;

[0112] Where, parameter n is the number of samples used for gradient estimation;

[0113] After obtaining the weights, the auxiliary coefficients are calculated based on formula (3) and the obtained weights. Specifically as follows:

[0114]

[0115] Among them, A k (i,j) is the activation value of node k at position (i,j) in the target layer;

[0116] Where Y(c) is the output score of the model for class c before softmax;

[0117] The ReLU function is primarily used to retain only the positive gradients that contribute to the calculation, while discarding the negative gradients.

[0118] in,

[0119] Based on this, after obtaining the weights and auxiliary coefficients, the location map L is calculated using the weights and auxiliary coefficients and formula (4), as follows:

[0120]

[0121] As one embodiment of the present invention, step S400 specifically includes:

[0122] The first additional feature obtained in step S200 is connected to one dimension of the input image, making the image size h*w*(d+1), which is then used for subsequent object detection.

[0123] The second additional feature obtained in step S300 is used as a mask to cover the channel dimension of the input original image, so that the shape of the input image becomes h*w*(d+1+1), which is used for subsequent target detection; by merging the mask into the channel dimension, non-pattern perception of the human brain is simulated.

[0124] In the above steps, w represents the image width, h represents the image height, and d represents the number of channels in the image.

[0125] Based on the above steps, predictive coding methods and non-modal perception methods can be used to simulate the visual feedback mechanism of the human brain. This allows for the quantification of the severity of the environment when performing target detection in harsh conditions, providing a basis for feedback adjustment. Furthermore, it enables accurate perception and prediction of occluded objects in the environment, improving the realism and accuracy of detection. This is especially applicable to autonomous driving, where it is more robust in adverse weather conditions.

[0126] Example 2

[0127] This embodiment is based on the same inventive concept as the target detection method based on brain-like feedback described in Embodiment 1, and provides a target detection system based on brain-like feedback, such as... Figure 4 As shown, it includes: a feedback network construction module, a brain-like feedback processing module, and a feature connection module;

[0128] The feedback network construction module is used to establish a brain-like feedback network based on the IECA module and the APC module;

[0129] In one embodiment of the present invention, the IECA module uses ResNet as the feature extractor; the last layer of the IECA module has a linear layer with an output dimension of 1; the loss function of the IECA model is L1 loss.

[0130] The neuromorphic feedback processing module is used to: evaluate the environmental severity of the input image using the IECA module of the neuromorphic feedback network when performing environmental recognition under severe weather conditions using a target detection model; generate a first additional feature for the target detection process based on the evaluation result; and select a region of interest in the input image using the APC module of the neuromorphic feedback network and perform nonmodal prediction completion processing to obtain a second additional feature for the target detection process.

[0131] As one embodiment of the present invention, before the IECA module using the brain-like feedback network performs environmental severity assessment processing on the input image, the process includes: the brain-like feedback processing module acquiring a dataset; the brain-like feedback processing module adding noise to the dataset to obtain a simulated foggy weather dataset; and the brain-like feedback processing module using the simulated foggy weather dataset to train the IECA module.

[0132] In one embodiment of the present invention, the APC module employing the neuromorphic feedback network selects a region of interest in the input image and performs non-modal prediction completion processing to obtain a second additional feature for the target detection process. This includes: the neuromorphic feedback processing module inputting the input image to the APC module; the neuromorphic feedback processing module outputting a one-hot vector through the backbone network of the APC module; the neuromorphic feedback processing module selecting the one-hot vector with the highest score as the activation class; the neuromorphic feedback processing module calculating the weight value of each pixel position in the image region of the activation class using a weight calculation formula; the neuromorphic feedback processing module calculating auxiliary coefficients based on the weight values ​​and coefficient calculation formula; the neuromorphic feedback processing module calculating a localization map based on the auxiliary coefficients and a localization map calculation formula; the neuromorphic feedback processing module retaining the obtained localization map as a grayscale image and adjusting the size of the grayscale image to fit the size of the input image; and the neuromorphic feedback processing module using the adjusted grayscale image as the second additional feature.

[0133] As one embodiment of the present invention, the weight calculation formula is: in: Let be the weight value of the activation class in the k-th layer at position (i,j); i∈[0,w] and j∈[0,w]; is the p-th partial derivative of the class score in the k-th layer with respect to the position (i,j) of class c; n is the number of samples used for gradient estimation.

[0134] As one embodiment of the present invention, the coefficient calculation formula is: in: A is the auxiliary coefficient; k (i,j) represents the activation value of node k at position (i,j) in the target layer; Y(c) represents the output score of the model for class c before softmax.

[0135] As one embodiment of the present invention, the formula for calculating the positioning map is: Where: L (c) (x,y) is the positioning map.

[0136] The feature connection module is used to input the input image, the first additional feature, and the second additional feature into the target detection model for target detection;

[0137] As one embodiment of the present invention, the step of inputting the input image, the first additional feature, and the second additional feature into the target detection model for target detection includes: the feature connection module connecting the first additional feature to one dimension of the input image, making the size of the input image h*w*(d+1), for target detection; the feature connection module using the second additional feature as a mask to cover the channel dimension of the input image, making the shape of the input image h*w*(d+1+1), for target detection; where w is the width of the input image, h represents the height of the input image, and d is the number of channels in the input image.

[0138] Example 3

[0139] This embodiment provides a computer-readable storage medium, including:

[0140] The storage medium is used to store computer software instructions for implementing the target detection method based on brain-like feedback as described in Embodiment 1 above. It includes a program for executing the target detection method based on brain-like feedback described above. Specifically, the executable program can be built into the target detection system based on brain-like feedback described in Embodiment 2. In this way, the target detection system based on brain-like feedback can implement the target detection method based on brain-like feedback described in Embodiment 1 by executing the built-in executable program.

[0141] Furthermore, the computer-readable storage medium in this embodiment can be any combination of one or more readable storage media, wherein the readable storage medium includes an electrical, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof.

[0142] Example 4

[0143] This embodiment provides an electronic device, such as... Figure 5 As shown, the electronic device may include: a processor 1501, a communication interface 1502, a memory 1503, and a communication bus 1504, wherein the processor 1501, the communication interface 1502, and the memory 1503 communicate with each other through the communication bus 1504.

[0144] Memory 1503 is used to store computer programs;

[0145] When the processor 1501 executes the computer program stored in the memory 1503, it implements the steps of the target detection method based on neuromorphic feedback described in Embodiment 1 above.

[0146] As one embodiment of the present invention, the communication bus mentioned in the terminal above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0147] As one embodiment of the present invention, the communication interface is used for communication between the aforementioned terminal and other devices.

[0148] In one embodiment of the present invention, the memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0149] As one embodiment of the present invention, the processor described above may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal 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.

[0150] Unlike existing technologies, the target detection method, system, device and medium based on brain-like feedback proposed in this application can simulate the human brain's visual feedback mechanism in target detection tasks under harsh environments based on the design of brain-like feedback networks. It can predict and encode targets under adverse environmental conditions and perform non-modal perception completion, thereby achieving more accurate target detection results, improving detection authenticity and robustness, and has high application value.

[0151] It should be understood that in the various embodiments of this document, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this document.

[0152] It should also be understood that, in the embodiments herein, the term "and / or" is merely a description of the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following associated objects have an "or" relationship.

[0153] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this document.

[0154] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0155] In the embodiments provided herein, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of 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. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, apparatuses, or units, or they may be electrical, mechanical, or other forms of connection.

[0156] 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 the embodiments described herein, depending on actual needs.

[0157] Furthermore, the functional units in the various embodiments of this document 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.

[0158] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this paper, 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.) to execute all or part of the steps of the methods described in the various embodiments of this paper. 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.

[0159] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A target detection method based on brain-like feedback, characterized in that, The method comprises the following steps: Feedback network building: A brain-like feedback network based on an IECA module and an APC module is established; Brain-like feedback processing: When a target detection model is used for environment recognition in bad weather: The IECA module using the brain-like feedback network performs environmental severity evaluation processing on the input image to obtain an evaluation result; and a first additional feature for a target detection process is generated based on the evaluation result; specifically, the IECA module uses ResNet as a feature extractor, the last layer of the IECA module has a linear layer to generate an output dimension of 1; the loss function of the IECA model selects L1 loss; after the IECA module evaluation, a corresponding severity value is obtained s The severity value is filled into a matrix with the same size as the input image as global information for subsequent target detection, simulating human feedback to judge global information, which is the first additional feature; the IECA module is an Initial Environmental Cognition Assessment module, and the APC module is an Amodal Predictive Completion module. The APC module using the brain-like feedback network selects a region of interest in the input image and performs non-modal prediction completion processing to obtain a second additional feature for the target detection process. Specifically, when performing APC analysis, the APC backbone network outputs a one-hot vector where C is the number of classes in the data set; after obtaining the output of the backbone network , the class with the highest score is selected as the active class c; wherein, ; the ArgMax function is used to select the class with the highest probability as the active class; then the weight of each pixel position in the image region of the active class is calculated using the weight calculation formula, the auxiliary coefficient is calculated based on the weight, and then the positioning map is calculated according to the auxiliary coefficient; the finally calculated positioning map is reserved as a grayscale image, and the size of the grayscale image is adjusted to adapt to the size of the original image x input into the target detection model; the size-adjusted grayscale image is used as the second additional feature; Feature connection: The input image, the first additional feature, and the second additional feature are input into the target detection model for target detection.

2. The target detection method based on brain-like feedback according to claim 1, wherein before the IECA module of the brain-like feedback network is used to perform environment severity evaluation processing on an input image, the method comprises the following steps: Obtaining a data set; Adding noise to the data set to obtain a simulated fog weather data set; Training the IECA module using the simulated fog weather data set.

3. The target detection method based on brain-like feedback according to claim 1, wherein the weight calculation formula is: wherein: n is the number of samples used for gradient estimation; ； 4. The target detection method based on brain-like feedback according to claim 3, wherein the coefficient calculation formula is: = ； ； the weight value of the activation class in the k-th layer at position ; , and ; w, input image width; the position of the class score in the k-th layer with respect to class c p-th order partial derivative of f with respect to x wherein: Ak is the activation value of node k in the kth layer (target layer) at pixel position (a, b); , where p is a pixel position; For a model's output score for class c before softmax activation, reflects the model's predicted confidence in that class. H is the height of the input image, and W is the width of the input image; The positioning map calculation formula is: ； 5. The target detection method based on brain-like feedback according to claim 1, wherein the inputting of the input image, the first additional feature, and the second additional feature into the target detection model for target detection comprises the following steps: wherein a is the auxiliary coefficient; Connecting the first additional feature to the one-dimensional input image, so that the size of the input image becomes h*w*(d+1), for the target detection; , is the activation value of node k in the target layer at position . , the output scores for c classes before softmax for the model; ; Covering the second additional feature as a mask in the channel dimension of the input image, so that the shape of the input image becomes h*w*(d+1+1), for the target detection; ； wherein: is the positioning map. wherein w is the width of the input image, h is the height of the input image, and d is the number of channels in the input image. comprise: a feedback network building module, a brain-like feedback processing module, and a feature connection module; The feedback network building module is configured to establish a brain-like feedback network based on an IECA module and an APC module; the IECA module is an Initial Environmental Cognition Assessment module, the APC module is an Amodal Predictive Completion module, and the IECA module and the APC module are used to realize a real visual feedback mechanism of simulating the human brain for target detection in bad weather; The brain-like feedback processing module is configured to, when a target detection model is used for environment recognition in bad weather: use the IECA module of the brain-like feedback network to perform environment severity evaluation processing on an input image to obtain an evaluation result; and generate a first additional feature for a target detection process based on the evaluation result.

6. A brain-like feedback based target detection system, characterized by, ​ ​ ​ ​ The brain-like feedback processing module selects an interest region in the input image by using an APC module of the brain-like feedback network, and performs a non-modal prediction completion process to obtain a second additional feature used in the target detection process; specifically, the IECA module uses a ResNet as a feature extractor, the last layer of the IECA module has a linear layer to generate an output dimension of 1, and the loss function of the IECA model selects an L1 loss; after the IECA module is evaluated, a corresponding severity value is obtained s The severity value is filled into a matrix with the same size as the input image as global information used in subsequent target detection, which simulates human feedback to judge global information, and the information is the first additional feature; specifically, when the APC analysis is performed, the APC backbone network outputs a one-hot vector where C is the number of classes in the data set; after the output of the backbone network is obtained , the class with the highest score is selected as the activated class c; wherein, The ArgMax function is used to select the class with the highest probability as the activated class; then the weight of each pixel position in the image region of the activated class is calculated by using a weight calculation formula, an auxiliary coefficient is calculated based on the weight, and a positioning map is calculated according to the auxiliary coefficient; the finally calculated positioning map is reserved as a gray image, the size of the gray image is adjusted to adapt to the size of the original image x input into the target detection model; and the gray image with the adjusted size is used as the second additional feature. The feature connection module is configured to input the input image, the first additional feature and the second additional feature into the target detection model for target detection.

7. A computer readable storage medium characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the target detection method based on the brain-like feedback according to any one of claims 1-5.

8. A computer device, comprising: The device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory can communicate with each other through the communication bus; wherein: The memory is configured to store a computer program. The processor is configured to execute the steps of the target detection method based on the brain-like feedback according to any one of claims 1-5 by running the program stored in the memory.

Images