A passage occupation recognition method, device, equipment and storage medium

Abstract

The application discloses a channel occupation recognition method and device, equipment and storage medium, and relates to the field of target recognition, and comprises the following steps: fusing a mask image created based on a to-be-detected channel region and an image of a current frame to obtain an image of interest; using a binary classification network to recognize the image of interest to obtain an image category, a classification confidence and a high-dimensional feature map of the current frame; determining whether the classification confidence is greater than a preset threshold; if yes, updating the high-dimensional feature map of the current frame to a feature map set of the corresponding category based on a buffer fault tolerance mechanism, and outputting the image category and the classification confidence; and if no, fusing a feature map set corresponding to the image category to obtain a matching template, comparing the high-dimensional feature map of the current frame with the matching template, and determining whether to output the image category and the classification confidence based on a comparison result. The application improves the accuracy of channel occupation recognition and reduces the false alarm rate by using feature clustering analysis for feature comparison and a buffer fault tolerance mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of target recognition, in particular to a passage occupation recognition method, device, equipment and storage medium. BACKGROUND

[0002] The fire-fighting passage is a life passage, and is in an idle state when not in use. Due to the compactness of urban space, the occupation and obstruction of the fire-fighting passage often occur. When a real fire occurs, the fire truck cannot arrive at the fire scene in time due to the occupation and obstruction of the fire-fighting passage, and thus human and property loss events frequently occur. Real-time monitoring and early warning of the occupation of the fire-fighting passage is a problem that needs to be solved in the current society. In view of this problem, some communities use manual on-site patrol or call indoor corridor monitoring to perform inspection. This method is time-consuming and laborious, and cannot timely find problems. Some communities install intelligent monitoring cameras to monitor the occurrence of events in real time. Although the intelligent monitoring algorithm needs to be further optimized to improve the accuracy and recall rate of the alarm, the intelligent monitoring algorithm needs to be further optimized to improve the accuracy and recall rate of the alarm.

[0003] In the prior art, a mask image is made for an image, and a binary image is obtained by graying. A foreground image is extracted based on a mixed Gaussian background modeling method. Then, a sliding window with overlap is used to uniformly sample the video image, and it is judged whether an object occupies the emergency passage by calculating whether the pixel values at the corresponding positions of the sampled images in the sliding window change. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a passage occupation recognition method, device, equipment and storage medium, which can extract the region of interest image from the intersection result, concentrate the network attention on the channel area to be detected, and improve the target diversity and universality in passage occupation recognition by using a binary classification network model. In addition, the feature comparison and buffer fault tolerance mechanism are used to further improve the accuracy of passage occupation recognition and reduce the false alarm rate. The specific scheme is as follows:

[0005] In the first aspect, the present application provides a passage occupation recognition method, comprising:

[0006] acquire a current frame image and create a mask image with the same size as the current frame image according to a channel region to be detected, perform intersection operation on the mask image and the current frame image, and cut out a region of interest image corresponding to the channel region to be detected from the intersection result;

[0007] perform channel occupancy recognition on the region of interest image by using a pre-trained binary classification network model to obtain a corresponding image category, an image classification confidence, and a current frame high-dimensional feature map, and determine whether the image classification confidence is greater than a preset confidence threshold;

[0008] If yes, update the current frame high-dimensional feature map to a high-dimensional feature map set of the corresponding category based on a buffer fault tolerance mechanism, and output the image category and the image classification confidence;

[0009] If no, perform fusion calculation on the high-dimensional feature map set corresponding to the image category to obtain a matching template, perform feature comparison between the current frame high-dimensional feature map and the matching template based on the comparison result, and determine whether to output the image category and the image classification confidence.

[0010] Optionally, the acquiring a current frame image and creating a mask image with the same size as the current frame image according to a channel region to be detected, performing intersection operation on the mask image and the current frame image, and cutting out a region of interest image corresponding to the channel region to be detected from the intersection result comprises:

[0011] acquire a current frame image and determine a maximum horizontal coordinate, a minimum horizontal coordinate, a maximum vertical coordinate, and a minimum vertical coordinate based on coordinates of each detection point in the channel region to be detected in the current frame image;

[0012] determine two corresponding coordinate points according to the minimum horizontal coordinate and the minimum vertical coordinate, and the maximum horizontal coordinate and the maximum vertical coordinate;

[0013] create a mask image with the same size as the current frame image according to the channel region to be detected, and perform intersection operation on the mask image and the current frame image to obtain an intersection region image;

[0014] determine a region cutting range based on the two coordinate points, and perform image cutting on the intersection region image based on the region cutting range to obtain a region of interest image corresponding to the channel region to be detected.

[0015] Optionally, before the performing channel occupancy recognition on the region of interest image by using a pre-trained binary classification network model, the method further comprises:

[0016] randomly selecting an interested background image from a background image library, and binarizing the interested background image to obtain a corresponding background mask image;

[0017] randomly selecting an initial foreground image from a foreground image library, and binarizing the initial foreground image to obtain an initial foreground mask image;

[0018] scaling the initial foreground image and the initial foreground mask image based on a preset proportion determination rule to obtain a corresponding scaled foreground image and a scaled foreground mask image;

[0019] randomly selecting a starting position from the interested background image according to a preset starting position range calculation rule, and obtaining a partial region image on the interested background image based on the width and height of the scaled foreground image, and then performing an intersection operation on the partial region image and the scaled foreground mask image to obtain a corresponding edge image;

[0020] adding the edge image and each pixel point of the scaled foreground image bit by bit to obtain a new foreground image, and pasting the new foreground image to the interested background image to obtain a fusion image;

[0021] performing an intersection operation on the fusion image and the background mask image to obtain a multiple random enhancement image;

[0022] constructing a training set based on the multiple random enhancement image, and training a binary classification network model using the training set to obtain a trained binary classification network model.

[0023] Optionally, the scaling the initial foreground image and the initial foreground mask image based on the preset proportion determination rule to obtain the corresponding scaled foreground image and the scaled foreground mask image comprises:

[0024] randomly generating an initial proportion value based on a preset proportion threshold range of the initial foreground image and the interested background image;

[0025] determining a target scaling ratio according to the initial proportion value, an original size ratio of the initial foreground image, and an original size ratio of the interested background image;

[0026] scaling the initial foreground image and the initial foreground mask image based on the target scaling ratio to obtain the corresponding scaled foreground image and the scaled foreground mask image.

[0027] Optionally, the step of fusing and calculating a matching template from the set of high-dimensional feature maps corresponding to the image category, comparing the current frame's high-dimensional feature map with the matching template, and determining whether to output the image category and the image classification confidence based on the comparison result includes:

[0028] A weighted fusion calculation is performed on the set of high-dimensional feature maps corresponding to the image category to obtain a matching template;

[0029] The high-dimensional feature map of the current frame is compared with the matching template to obtain the corresponding difference value;

[0030] Determine whether the difference value is less than a preset difference threshold. If it is less than the threshold, output the image category and the image classification confidence score.

[0031] Optionally, the step of updating the current frame's high-dimensional feature map to the corresponding category's high-dimensional feature map set based on the buffer fault-tolerance mechanism, and outputting the image category and the image classification confidence score, includes:

[0032] Determine if the buffer set is in a direct cache state;

[0033] If not, then determine whether the category corresponding to the high-dimensional feature map of the current frame is the same as the category corresponding to the high-dimensional feature map of the previous frame;

[0034] If they are the same, the high-dimensional feature map of the current frame is updated to the high-dimensional feature map set of the corresponding category, and the image category and the image classification confidence are output.

[0035] If they are different, the high-dimensional feature map of the current frame is updated to the buffer set, and the buffer set is marked as the direct cache state.

[0036] Optionally, after determining whether the state of the buffer set is a direct cache state, the method further includes:

[0037] If so, the high-dimensional feature map of the current frame is directly cached into the buffer set;

[0038] When the number of high-dimensional feature maps in the buffer set reaches a preset number, a category analysis is performed on the preset number of high-dimensional feature maps in the buffer set to obtain classification results, and the target classification result with the largest proportion is selected from the classification results.

[0039] When the proportion corresponding to the target classification result exceeds a preset proportion threshold, the high-dimensional feature maps corresponding to the category of the target classification result in the buffer set are updated to the high-dimensional feature map set of the corresponding category, and the buffer set is cleared and marked as the end of analysis.

[0040] Secondly, this application provides a channel occupancy identification device, comprising:

[0041] The image of interest extraction module is used to acquire the current frame image and create a mask image of the same size as the current frame image based on the channel region to be detected, perform an intersection operation on the mask image and the current frame image, and extract the region of interest image corresponding to the channel region to be detected from the intersection result;

[0042] The confidence judgment module is used to perform channel occupancy recognition on the region of interest image using a pre-trained binary classification network model to obtain the corresponding image category, image classification confidence and current frame high-dimensional feature map, and to determine whether the image classification confidence is greater than a preset confidence threshold.

[0043] The set update module is used to update the current frame high-dimensional feature map to the high-dimensional feature map set of the corresponding category based on the buffer fault tolerance mechanism if the value is greater than the set, and output the image category and the image classification confidence.

[0044] The feature comparison module is used to perform a fusion calculation on the set of high-dimensional feature maps corresponding to the image category if the value is not greater than the specified value, to obtain a matching template, and to compare the current frame's high-dimensional feature map with the matching template. Based on the comparison result, it determines whether to output the image category and the image classification confidence.

[0045] Thirdly, this application provides an electronic device, comprising:

[0046] Memory, used to store computer programs;

[0047] A processor is used to execute the computer program to implement the aforementioned channel occupancy identification method.

[0048] Fourthly, this application provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the aforementioned channel occupancy identification method.

[0049] In this application, the current frame image is acquired, and a mask image of the same size as the current frame image is created based on the channel region to be detected. The intersection of the mask image and the current frame image is performed, and the region of interest image corresponding to the channel region to be detected is extracted from the intersection result. A pre-trained binary classification network model is used to identify channel occupancy in the region of interest image to obtain the corresponding image category, image classification confidence, and current frame high-dimensional feature map. It is then determined whether the image classification confidence is greater than a preset confidence threshold. If it is greater, the current frame high-dimensional feature map is updated to the corresponding category's high-dimensional feature map set based on a buffer fault tolerance mechanism, and the image category and the image classification confidence are output. If it is not greater, the high-dimensional feature map set corresponding to the image category is fused to obtain a matching template, and the current frame high-dimensional feature map is compared with the matching template. Based on the comparison result, it is determined whether to output the image category and the image classification confidence. Therefore, this application effectively removes background interference by extracting the region of interest image from the intersection of the mask image and the current frame image, thus focusing the attention of the binary classification network model on the channel to be detected. Furthermore, by employing a binary classification network model to identify channel occupancy in the region of interest image, all foreground objects occupying channels are grouped into one category, solving the problem of foreground object diversity and avoiding the need to collect a large number of different foreground training samples in the early stages of model training, thereby improving the versatility of channel occupancy identification. Finally, by using feature clustering analysis for feature comparison and employing a buffer fault-tolerance mechanism to filter uncertain channel occupancy identification results, the accuracy of channel occupancy identification is further improved, and the false alarm rate is reduced. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0051] Figure 1 This is a flowchart of a channel occupancy identification method disclosed in this application;

[0052] Figure 2 This is a schematic diagram of a region of interest image disclosed in this application;

[0053] Figure 3 This application discloses a flowchart for channel occupancy identification.

[0054] Figure 4 This is a flowchart of a specific channel occupancy identification method disclosed in this application;

[0055] Figure 5 This application discloses a flowchart for determining multiple randomized image enhancements.

[0056] Figure 6 This is a schematic diagram of a multiple randomized image enhancement method disclosed in this application;

[0057] Figure 7 This is a flowchart of a specific channel occupancy identification method disclosed in this application;

[0058] Figure 8 This application discloses a flowchart of a buffer fault-tolerance mechanism.

[0059] Figure 9 This is a schematic diagram of the structure of a channel occupancy identification device disclosed in this application;

[0060] Figure 10 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Currently, the obstruction and blockage of fire lanes prevents fire trucks from reaching fire scenes in a timely manner, leading to frequent incidents of personal injury and property damage. Real-time monitoring and early warning of fire lane obstruction is a pressing issue that needs to be addressed. To this end, this application provides a method for identifying fire lane obstruction. By extracting the region of interest from the image, the network's attention is focused on the area to be detected. A binary classification network model is employed to improve the target diversity and versatility of fire lane obstruction identification. Furthermore, feature clustering analysis is used for feature comparison and a buffering fault-tolerance mechanism to further improve the accuracy of fire lane obstruction identification and reduce the false alarm rate.

[0063] See Figure 1 As shown, an embodiment of the present invention discloses a channel occupancy identification method, including:

[0064] Step S11: Obtain the current frame image and create a mask image with the same size as the current frame image based on the channel region to be detected. Perform an intersection operation on the mask image and the current frame image, and extract the region of interest image corresponding to the channel region to be detected from the intersection result.

[0065] In this embodiment, the current frame image is acquired, and the maximum x-coordinate, minimum x-coordinate, maximum y-coordinate, and minimum y-coordinate are determined based on the coordinates of each detection point within the target channel region in the current frame image. Two corresponding coordinate points are determined based on the minimum x-coordinate and minimum y-coordinate, and the maximum x-coordinate and maximum y-coordinate. A mask image of the same size as the current frame image is created based on the target channel region, and an intersection operation is performed between the mask image and the current frame image to obtain an intersection region image. A region extraction range is determined based on the two coordinate points, and image extraction is performed on the intersection region image based on the region extraction range to obtain a region of interest image corresponding to the target channel region. It is understood that, as... Figure 2 As shown, the acquired original image is the current frame image, and the area within the red box in the current frame image is the region to be detected. Based on the coordinates of each detection point within the region to be detected (P1, P2, ..., P...),... i Get the maximum x-coordinate among all coordinates. max Minimum value of x-axis X min Maximum value of the ordinate Y max and the minimum value of the ordinate Y min The specific calculation formula is as follows:

[0066]

[0067] Based on the maximum x-coordinate X max and the maximum value of the ordinate Y max and the minimum value of the x-axis X min and the minimum value of the ordinate Y min Obtain the two corresponding coordinate points, which are the coordinates of the top left corner point P. top and the coordinates of the lower right corner point P bottom Then, a mask image of the same size as the current frame image is created based on the region to be detected, where the value within the region to be detected is 255, and the value in other regions is 0. The mask image and the current frame image are intersected to extract the region of interest, resulting in the intersection image. This is based on the coordinates of the top-left corner point P. top and the coordinates of the lower right corner point P bottom The region extraction range is determined in the intersection region image, and the region of interest is extracted from the intersection region image based on the region extraction range to form a miniature image. The final miniature image of the region of interest is shown below. Figure 2 As shown. In this way, by extracting the region of interest from the intersection of the mask image and the current frame image, the network's attention is focused on the channel region to be detected in the region of interest, and the interference of redundant background can be effectively removed.

[0068] Step S12: Use a pre-trained binary classification network model to perform channel occupancy recognition on the region of interest image to obtain the corresponding image category, image classification confidence and current frame high-dimensional feature map, and determine whether the image classification confidence is greater than a preset confidence threshold.

[0069] In this embodiment, considering that in real-world channel occupancy recognition algorithm applications, the foreground objects occupying a channel are of various types, resulting in extremely high sample complexity and making it impossible to exhaustively enumerate all foreground targets, this application groups all foreground objects occupying a channel into one category and employs a binary classification network model for channel occupancy recognition. This binary classification network model designs the recognition results into two categories: one for the channel itself and the other for objects occupying the channel. Specifically, as shown... Figure 3 As shown, after acquiring the region of interest (ROI) image, the ROI image is input into a pre-trained binary classification network model. This model is used to identify channel occupancy, obtaining the corresponding image category, image classification confidence score, and high-dimensional feature map of the current frame. The image classification confidence score is then compared with a pre-set confidence threshold to perform an initial filtering of the channel occupancy identification results. In this way, by employing a binary classification network model, this application groups all foreground objects occupying channels into one category, improving the versatility of channel occupancy identification and its adaptability to multiple scenarios.

[0070] Step S13: If the value is greater than the specified value, the high-dimensional feature map of the current frame is updated to the high-dimensional feature map set of the corresponding category based on the buffer fault tolerance mechanism, and the image category and the image classification confidence score are output.

[0071] In this embodiment, as Figure 3 As shown, if the image classification confidence score is greater than a preset confidence threshold, the current frame's high-dimensional feature map will be updated to the corresponding category's high-dimensional feature map set based on a buffer fault tolerance mechanism, and the image category and image classification confidence score will be output. For example, when the image category is the channel category and the image classification confidence score is greater than the preset confidence threshold TR, the current frame's high-dimensional feature map will be updated to the channel category's high-dimensional feature map set BB based on the buffer fault tolerance mechanism; when the image category is the channel-occupied object category and the image classification confidence score is greater than the preset confidence threshold TR, the current frame's high-dimensional feature map will be updated to the channel-occupied object category's high-dimensional feature map set OB based on the buffer fault tolerance mechanism, as specifically shown below:

[0072]

[0073] Where I(x) i f(x) represents the high-dimensional feature map of the current frame, BB represents the set of high-dimensional feature maps for each channel category, and f(x) represents the high-dimensional feature map of the current frame. i T(x) represents the image category.i ) represents the image classification confidence, TR represents the preset confidence threshold, 0 represents the channel category, 1 represents the channel has an object category, and OB represents the set of high-dimensional feature maps of the channel has an object category.

[0074] In this embodiment, considering that the images identified in the application scenarios of the channel occupancy recognition algorithm often do not change much, but changes in light, water accumulation, pedestrian interference, etc. can interfere with the recognition results, this application designs a buffer fault tolerance mechanism. The buffer set is used to process the high-dimensional feature map of the current frame that has interference, to prevent directly updating the high-dimensional feature map of the current frame with uncertain recognition results to the high-dimensional feature map set of the corresponding category, and to avoid interference with the matching results obtained based on the high-dimensional feature map set.

[0075] Step S14: If the value is not greater than the specified value, the high-dimensional feature map set corresponding to the image category is fused and calculated to obtain a matching template. The high-dimensional feature map of the current frame is compared with the matching template. Based on the comparison result, it is determined whether to output the image category and the image classification confidence.

[0076] In this embodiment, as Figure 3 As shown, if the image classification confidence is less than or equal to a preset confidence threshold, a weighted fusion calculation is performed on the high-dimensional feature map set corresponding to the image category to obtain a matching template. Then, the high-dimensional feature map of the current frame is compared with the matching template to obtain the corresponding difference value. Finally, it is determined whether the difference value is less than a preset difference threshold. If it is less, the image category and image classification confidence are output. It can be understood that the matching template is generated by a weighted fusion of the high-dimensional feature map set of the corresponding category. The higher the image classification confidence, the larger its proportion, and the closer it is to the current frame. The specific weighted fusion calculation formula is as follows:

[0077]

[0078] Where I(x) m I(x) represents the matching template generated from a set of high-dimensional feature maps corresponding to the image category. i T(x) represents the high-dimensional feature map of the current frame. i ) represents the confidence level for image classification. The smaller the value of 'i', the closer it is to the current frame, and the greater its weight percentage.

[0079] In this embodiment, after obtaining the matching template, the similarity between the current frame's high-dimensional feature map and the matching template is compared to obtain the corresponding difference value. Then, the difference value is compared with a preset difference threshold. When the difference value is less than the preset difference threshold, the image category and image classification confidence score are output. In this way, this application uses feature clustering analysis to compare the features of the current frame's high-dimensional feature map with the matching template, thereby reconfirming the classification results where the image classification confidence score is less than the preset confidence threshold, avoiding directly outputting the image category and image classification confidence score where the channel occupancy recognition result is uncertain.

[0080] Therefore, this application effectively removes background interference by extracting the region of interest image from the intersection of the mask image and the current frame image, thus focusing the attention of the binary classification network model on the channel to be detected. Furthermore, by employing a binary classification network model to identify channel occupancy in the region of interest image, all foreground objects occupying channels are grouped into one category, solving the problem of foreground object diversity and avoiding the need to collect a large number of different foreground training samples in the early stages of model training, thereby improving the versatility of channel occupancy identification. Finally, by using feature clustering analysis for feature comparison and employing a buffer fault-tolerance mechanism to filter uncertain channel occupancy identification results, the accuracy of channel occupancy identification is further improved, and the false alarm rate is reduced.

[0081] As described in the previous embodiment, this application uses a pre-trained binary classification network model to perform channel occupancy recognition on the region of interest image. Therefore, this application will now elaborate on how to train the binary classification network model. See [link to previous document]. Figure 4 As shown, this embodiment of the invention discloses a process for training a binary classification network model, including:

[0082] Step S21: Randomly select a background image of interest from the background image library, and binarize the background image of interest to obtain the corresponding background mask image.

[0083] In this embodiment, several images of interest from different channels are acquired to form a background image library. Further, as... Figure 5 As shown, a background image of interest BI is randomly selected from the background image library, and the background image of interest BI is binarized to obtain the background mask image BM corresponding to the background image of interest BI.

[0084] Step S22: Randomly select an initial foreground image from the foreground image library and binarize the initial foreground image to obtain an initial foreground mask image.

[0085] In this embodiment, due to the diversity and irregular boundaries of objects in the foreground image, to ensure the effectiveness of model training, this application first needs to calibrate the objects in the foreground image, completely segment the boundaries of the objects, and manually form corresponding mask images. Finally, a foreground image library is constructed based on the foreground image and the corresponding mask images. Further, as... Figure 5 As shown, an initial foreground image OI is randomly selected from the foreground image library, and the initial foreground image OI is binarized to obtain the initial foreground mask image OM corresponding to the initial foreground image OI.

[0086] Step S23: Scale the initial foreground image and the initial foreground mask image according to the preset ratio determination rule to obtain the corresponding scaled foreground image and scaled foreground mask image.

[0087] In this embodiment, scaling the initial foreground image and the initial foreground mask image based on a preset ratio determination rule to obtain corresponding scaled foreground images and scaled foreground mask images may include: randomly generating an initial ratio value based on a preset ratio threshold range between the initial foreground image and the background of interest; determining a target scaling ratio based on the initial ratio value, the original size ratio of the initial foreground image, and the original size ratio of the background of interest; and scaling the initial foreground image and the initial foreground mask image based on the target scaling ratio to obtain corresponding scaled foreground images and scaled foreground mask images. It is understood that, as... Figure 5 As shown, the initial threshold range for the proportion of the foreground image OI to the background image of interest BI is pre-set to be (T). min T max An initial scaling factor Tt is randomly generated based on the preset scaling factor threshold range. According to the initial scaling factor Tt and the original size ratio of the initial foreground image OI to the background image of interest BI, the final target scaling factor is obtained. Finally, the initial foreground image OI and the initial foreground mask image OM are scaled according to the target scaling factor to obtain the scaled foreground image OI and the scaled foreground mask image OM. The scaled foreground image OI is shown below. Figure 6 As shown.

[0088] Step S24: Randomly select a starting position from the background image of interest according to the preset starting position range calculation rules, and obtain a partial area image on the background image of interest based on the width and height of the scaled foreground image. Then, perform an intersection operation between the partial area image and the scaled foreground mask image to obtain the corresponding edge image.

[0089] In this embodiment, the formula for calculating the preset starting position range is as follows:

[0090]

[0091] Among them, OI width Indicates the width of the scaled foreground image, OI height BI represents the height of the scaled foreground image. width BI represents the width of the background image of interest. height The height of the background image of interest is represented by BI(x,y), and the pixel value at position (x,y) in the background image of interest is represented by BI(x,y). Further, as... Figure 5 As shown, after obtaining the starting position range, a starting position (x, y) is randomly selected in the background image BI of interest based on the starting position range, and the width OI of the scaled foreground image is used as the starting position. width and the height OI of the scaled foreground image height A partial region of image [y:y+OI] is identified in the background image of interest (BI). height ,x:x+OI width The intersection operation is performed between a portion of the image and the scaled foreground mask image OM to obtain the corresponding edge image RI.

[0092] Step S25: Add each pixel of the edge image to the scaled foreground image bit by bit to obtain a new foreground image, and then paste the new foreground image onto the background image of interest to obtain a fused image.

[0093] In this embodiment, as Figure 5 As shown, the edge image RI and the scaled foreground image OI are added pixel by pixel to complete the merging and obtain the merged image. Then, the real background of the background image BI of interest is fused into the merged image to generate a new background image OIN, where the new background image OIN is as follows. Figure 6 As shown. Further, the new background image OIN is mapped onto the background image of interest BI to synthesize... Figure 6 The fused image BX is shown.

[0094] Step S26: Perform an intersection operation on the fused image and the background mask image to obtain multiple randomly enhanced images.

[0095] In this embodiment, as Figure 5 and Figure 6 As shown, the intersection operation is performed between the fused image BX and the background mask image BM to cut out the excess image beyond the background mask image BM, completing the realistic texture mapping and obtaining the multiple randomly enhanced image FI, which is... Figure 6 The final image FI is shown.

[0096] Step S27: Construct a training set based on the multiple randomly enhanced images, and use the training set to train the binary classification network model to obtain the trained binary classification network model.

[0097] In this embodiment, considering that channel occupancy is relatively rare in reality and sample collection is quite difficult, this application designs a multiple random image enhancement algorithm suitable for channel occupancy identification to enrich sample diversity. A model training set is constructed using the multiple randomly enhanced images obtained based on the multiple random image enhancement algorithm, and the training set is used to train a binary classification network model to obtain a trained binary classification network model. The multiple randomness of the multiple random image enhancement algorithm in this embodiment is reflected in the randomness of the selection of foreground and background images, the randomness of the initial proportion value generated when scaling the foreground image, and the randomness of selecting the starting position in the background image of interest. This ensures the randomness, diversity, and realism of the multiple randomly enhanced images obtained through texturing.

[0098] Therefore, this application enriches the diversity of training samples by adopting a multi-random image enhancement algorithm suitable for channel occupancy recognition, ensuring the randomness and authenticity of the multi-random enhanced images obtained by mapping; at the same time, by using the training set constructed based on the multi-random enhanced images to train the binary classification network model, the binary classification network model can learn and extract image features more effectively, thereby improving the accuracy of channel occupancy recognition.

[0099] Based on the previous embodiment, this application describes the process of training a binary classification network model and the overall process of channel occupancy recognition for a region of interest image. Next, this application will elaborate on the buffer fault-tolerance mechanism. See [link to previous document]. Figure 7 As shown, this embodiment of the invention discloses the working process of a buffer fault-tolerance mechanism, including:

[0100] Step S31: Determine whether the state of the buffer set is direct caching. If so, directly cache the high-dimensional feature map of the current frame into the buffer set.

[0101] In this embodiment, when the image classification confidence score is greater than a preset confidence threshold, to avoid updating the high-confidence results of the current frame detection results that may contain false positives due to factors such as lighting, pedestrian interference, or road surface water to the corresponding category's high-dimensional feature map set, a buffer fault-tolerance mechanism is added. This buffer set is used to further process the high-dimensional feature maps of the current frame that contain interference. Furthermore, as... Figure 8As shown, the process first checks if the buffer set is in the direct caching state (true). If it is, the current frame's high-dimensional feature map is directly updated to the buffer set. If not, meaning the buffer set is in the end-of-analysis state (false), the process further checks if the category corresponding to the current frame's high-dimensional feature map is the same as the category corresponding to the previous frame's high-dimensional feature map. It should be noted that the "previous frame's high-dimensional feature map" refers to the image preceding the current frame's high-dimensional feature map during the actual acquisition time. If the category corresponding to the current frame's high-dimensional feature map is the same as the category corresponding to the previous frame's high-dimensional feature map, the current frame's high-dimensional feature map is updated to the corresponding category's high-dimensional feature map set, and the image category and classification confidence score are output.

[0102] In this embodiment, if the category corresponding to the high-dimensional feature map of the current frame is different from the category corresponding to the high-dimensional feature map of the previous frame, the high-dimensional feature map of the current frame is updated to the buffer set, and the buffer set is marked as a direct cache state (true). At this time, the high-dimensional feature map of the current frame is the first high-dimensional feature map stored in the buffer set, and the buffer set is not analyzed temporarily. High-dimensional feature maps with a classification confidence level greater than a preset confidence threshold for a preset number of consecutive frames following the high-dimensional feature map of the current frame are cached to the buffer set; that is, when the number of high-dimensional feature maps in the buffer set reaches a preset number, the buffer set is then analyzed. For example, if the high-dimensional feature map of the current frame is the high-dimensional feature map of the second frame, and the category corresponding to the high-dimensional feature map of the second frame is different from the category corresponding to the high-dimensional feature map of the previous frame, then after caching the high-dimensional feature map of the second frame to the buffer set, the buffer set is marked as a direct cache state true. At this time, if the image classification confidence of the high-dimensional feature map of the third frame is greater than the preset confidence threshold, since the state of the buffer set is a direct cache state, the high-dimensional feature map of the third frame is directly cached to the buffer set. This process continues until the number of high-dimensional feature maps in the buffer set reaches the preset number.

[0103] Step S32: When the number of high-dimensional feature maps in the buffer set reaches a preset number, perform category analysis on the preset number of high-dimensional feature maps in the buffer set to obtain classification results, and select the target classification result with the largest proportion from the classification results.

[0104] In this embodiment, as Figure 8As shown, when the number of high-dimensional feature maps in the buffer set reaches a preset number of 5, category analysis can begin on the 5 high-dimensional feature maps in the buffer set, and the target classification result with the largest proportion in the classification results will be obtained. The preset number can be set according to the user's actual needs. If the number of high-dimensional feature maps in the buffer set has not yet reached the preset number, the buffer set will not be analyzed temporarily to avoid the problem of low reliability in classification results caused by analyzing the buffer set when the number of high-dimensional feature maps is small.

[0105] Step S33: When the proportion corresponding to the target classification result exceeds the preset proportion threshold, update the high-dimensional feature maps in the buffer set corresponding to the category of the target classification result to the high-dimensional feature map set of the corresponding category, clear the buffer set and mark the buffer set as the end of analysis state.

[0106] In this embodiment, as Figure 8 As shown, when the proportion corresponding to the target classification result exceeds the preset proportion threshold, several high-dimensional feature maps corresponding to the category of the target classification result in the buffer set are updated to the corresponding category's high-dimensional feature map set. The buffer set is then cleared and set to the false label, meaning the buffer set can start saving high-dimensional feature maps again and is in a state of analysis termination. The preset proportion threshold can be set according to the user's actual proportion requirements and should be greater than or equal to 0.5.

[0107] Therefore, this application, when the image classification confidence level is greater than the preset confidence threshold, adds a buffer fault tolerance mechanism. When the category of the channel occupancy recognition result suddenly changes, the current frame's high-dimensional feature map is cached in a buffer set for buffering. After performing feature clustering analysis on a preset number of high-dimensional feature maps stored in the buffer set, it is determined whether to update to the high-dimensional feature map set of the corresponding category. This avoids directly updating the high-dimensional feature map set of the corresponding category with interfering high-dimensional feature maps, thereby improving the accuracy of channel occupancy recognition and reducing the false alarm rate.

[0108] See Figure 9 As shown, this embodiment of the invention also discloses a channel occupancy identification device, comprising:

[0109] The image of interest extraction module 11 is used to acquire the current frame image and create a mask image of the same size as the current frame image according to the channel region to be detected, perform an intersection operation on the mask image and the current frame image, and extract the region of interest image corresponding to the channel region to be detected from the intersection result;

[0110] The confidence judgment module 12 is used to perform channel occupancy recognition on the region of interest image using a pre-trained binary classification network model to obtain the corresponding image category, image classification confidence and current frame high-dimensional feature map, and to determine whether the image classification confidence is greater than a preset confidence threshold.

[0111] The set update module 13 is used to update the current frame high-dimensional feature map to the high-dimensional feature map set of the corresponding category based on the buffer fault tolerance mechanism if the value is greater than the set, and output the image category and the image classification confidence.

[0112] The feature comparison module 14 is used to perform a fusion calculation on the set of high-dimensional feature maps corresponding to the image category if the value is not greater than the specified value, to obtain a matching template, and to compare the current frame high-dimensional feature map with the matching template, and to determine whether to output the image category and the image classification confidence based on the comparison result.

[0113] Therefore, this application effectively removes background interference by extracting the region of interest image from the intersection of the mask image and the current frame image, thus focusing the attention of the binary classification network model on the channel to be detected. Furthermore, by employing a binary classification network model to identify channel occupancy in the region of interest image, all foreground objects occupying channels are grouped into one category, solving the problem of foreground object diversity and avoiding the need to collect a large number of different foreground training samples in the early stages of model training, thereby improving the versatility of channel occupancy identification. Finally, by using feature clustering analysis for feature comparison and employing a buffer fault-tolerance mechanism to filter uncertain channel occupancy identification results, the accuracy of channel occupancy identification is further improved, and the false alarm rate is reduced.

[0114] In some specific embodiments, the image of interest extraction module 11 may specifically include:

[0115] The horizontal and vertical coordinate determination unit is used to acquire the current frame image and determine the maximum horizontal coordinate, minimum horizontal coordinate, maximum vertical coordinate, and minimum vertical coordinate based on the coordinates of each detection point in the detection channel region of the current frame image.

[0116] The coordinate point determination unit is used to determine two corresponding coordinate points based on the minimum abscissa and the minimum ordinate, and the maximum abscissa and the maximum ordinate;

[0117] The image intersection unit is used to create a mask image with the same size as the current frame image based on the channel region to be detected, and to perform an intersection operation on the mask image and the current frame image to obtain an intersection region image;

[0118] The image extraction unit is used to determine the region extraction range based on the two coordinate points, and to extract the image of the intersection region based on the region extraction range, so as to obtain the region of interest image corresponding to the channel region to be detected.

[0119] In some specific embodiments, the channel occupancy identification device may further include:

[0120] The background image acquisition unit is used to randomly select a background image of interest from the background image library and binarize the background image of interest to obtain the corresponding background mask image.

[0121] The foreground image acquisition unit is used to randomly select an initial foreground image from the foreground image library and binarize the initial foreground image to obtain an initial foreground mask image.

[0122] The foreground image scaling submodule is used to scale the initial foreground image and the initial foreground mask image respectively based on a preset ratio determination rule to obtain the corresponding scaled foreground image and scaled foreground mask image.

[0123] The edge image determination unit is used to randomly select a starting position from the background image of interest according to a preset starting position range calculation rule, and obtain a partial region image on the background image of interest based on the width and height of the scaled foreground image. Then, the partial region image is intersected with the scaled foreground mask image to obtain the corresponding edge image.

[0124] An image mapping unit is used to add each pixel of the edge image to the scaled foreground image bit by bit to obtain a new foreground image, and then map the new foreground image onto the background image of interest to obtain a fused image;

[0125] An enhanced image acquisition unit is used to perform an intersection operation between the fused image and the background mask image to obtain multiple random enhanced images;

[0126] The model training unit is used to construct a training set based on the multiple randomly enhanced images, and to train the binary classification network model using the training set to obtain a trained binary classification network model.

[0127] In some specific embodiments, the foreground image scaling submodule may specifically include:

[0128] The proportion value generation unit is used to randomly generate an initial proportion value based on a preset proportion threshold range between the initial foreground image and the background image of interest;

[0129] A scaling ratio determination unit is used to determine a target scaling ratio based on the initial ratio value, the original size ratio of the initial foreground image, and the original size ratio of the background image of interest.

[0130] An image scaling unit is used to scale the initial foreground image and the initial foreground mask image respectively based on the target scaling ratio to obtain the corresponding scaled foreground image and scaled foreground mask image.

[0131] In some specific embodiments, the feature comparison module 14 may specifically include:

[0132] The matching template acquisition unit is used to perform weighted fusion calculation on the set of high-dimensional feature maps corresponding to the image category to obtain the matching template;

[0133] The feature comparison unit is used to compare the current frame high-dimensional feature map with the matching template to obtain the corresponding difference value;

[0134] The difference value determination unit is used to determine whether the difference value is less than a preset difference threshold. If it is less than the threshold, the image category and the image classification confidence level are output.

[0135] In some specific embodiments, the set update module 13 may specifically include:

[0136] The set state judgment unit is used to determine whether the state of the buffer set is a direct cache state;

[0137] The category determination unit is used to determine whether the category corresponding to the high-dimensional feature map of the current frame is the same as the category corresponding to the high-dimensional feature map of the previous frame if no.

[0138] The first set update unit is used to update the current frame high-dimensional feature map to the high-dimensional feature map set of the corresponding category if they are the same, and output the image category and the image classification confidence.

[0139] The set marking unit is used to update the current frame high-dimensional feature map to the buffer set if they are different, and to mark the buffer set as the direct cache state.

[0140] In some specific embodiments, the channel occupancy identification device may further include:

[0141] A feature map caching unit is used to directly cache the high-dimensional feature map of the current frame into the buffer set if the condition is met.

[0142] The category analysis unit is used to perform category analysis on the preset number of high-dimensional feature maps in the buffer set when the number of high-dimensional feature maps in the buffer set reaches a preset number, so as to obtain classification results and filter out the target classification result with the largest proportion from the classification results.

[0143] The second set update unit is used to update several high-dimensional feature maps in the buffer set corresponding to the category of the target classification result to the high-dimensional feature map set of the corresponding category when the proportion corresponding to the target classification result exceeds a preset proportion threshold, and to clear the buffer set and mark the buffer set as the end of analysis state.

[0144] Furthermore, embodiments of this application also disclose an electronic device, Figure 10 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0145] Figure 10 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the channel occupancy identification method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0146] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0147] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0148] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the channel occupancy identification method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program capable of performing other specific tasks.

[0149] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned channel occupancy identification method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0150] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0151] Those skilled in the art will further 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 application.

[0152] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0153] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0154] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method of identifying passage occupancy, characterized by, The method comprises the following steps: obtaining a current frame image and creating a mask image with the same size as the current frame image according to a channel region to be detected, performing intersection operation on the mask image and the current frame image, and cutting out a region of interest image corresponding to the channel region to be detected from the intersection result; using a pre-trained binary classification network model to perform channel occupancy recognition on the region of interest image to obtain a corresponding image category, an image classification confidence and a current frame high-dimensional feature map, and determining whether the image classification confidence is greater than a pre-set confidence threshold; if yes, updating the current frame high-dimensional feature map to a high-dimensional feature map set of the corresponding category based on a buffer fault tolerance mechanism, and outputting the image category and the image classification confidence; if no, performing fusion calculation on the high-dimensional feature map set corresponding to the image category to obtain a matching template, and performing feature comparison between the current frame high-dimensional feature map and the matching template based on the comparison result to determine whether to output the image category and the image classification confidence; wherein the updating of the current frame high-dimensional feature map to the high-dimensional feature map set of the corresponding category based on the buffer fault tolerance mechanism comprises: when the image category suddenly changes, the current frame high-dimensional feature map is buffered to a buffer set through the buffer fault tolerance mechanism, and the high-dimensional feature maps with image classification confidence greater than the pre-set confidence threshold in the subsequent pre-set number of frames are buffered to the buffer set, so that the number of high-dimensional feature maps in the buffer set reaches a pre-set number, and after category analysis is performed on the pre-set number of high-dimensional feature maps stored in the buffer set, it is determined whether to update to the high-dimensional feature map set of the corresponding category.

2. The method of aisle occupancy recognition of claim 1, wherein, The method comprises the following steps: obtaining a current frame image and creating a mask image with the same size as the current frame image according to a channel region to be detected, performing intersection operation on the mask image and the current frame image, and cutting out a region of interest image corresponding to the channel region to be detected from the intersection result; obtaining a current frame image and determining the maximum horizontal coordinate, the minimum horizontal coordinate, the maximum vertical coordinate and the minimum vertical coordinate based on the coordinates of each detection point in the channel region to be detected in the current frame image; determining two corresponding coordinate points according to the minimum horizontal coordinate and the minimum vertical coordinate and the maximum horizontal coordinate and the maximum vertical coordinate; creating a mask image with the same size as the current frame image according to the channel region to be detected, and performing intersection operation on the mask image and the current frame image to obtain an intersection region image; 3. The method of aisle occupancy recognition of claim 1, wherein, determining a region cutting range based on the two coordinate points, and performing image cutting on the intersection region image based on the region cutting range to obtain a region of interest image corresponding to the channel region to be detected. Before the channel occupancy recognition on the region of interest image using the pre-trained binary classification network model, the method further comprises the following steps: randomly selecting a background image of interest from a background image library, and performing binaryzation on the background image of interest to obtain a corresponding background mask image; randomly select an initial foreground image from the foreground image library, and binarize the initial foreground image to obtain an initial foreground mask image; scaling the initial foreground image and the initial foreground mask image based on a preset proportion determination rule to obtain a corresponding scaled foreground image and a scaled foreground mask image; randomly selecting a starting position from the background image of interest according to a preset starting position range calculation rule, and obtaining a partial region image on the background image of interest based on the width and height of the scaled foreground image, then performing intersection operation on the partial region image and the scaled foreground mask image to obtain a corresponding edge image; adding the edge image and each pixel point of the scaled foreground image bit by bit to obtain a new foreground image, and pasting the new foreground image to the background image of interest to obtain a fusion image; performing intersection operation on the fusion image and the background mask image to obtain a multiple random enhancement image; constructing a training set based on the multiple random enhancement image, and training a binary classification network model using the training set to obtain a trained binary classification network model.

4. The method of aisle occupancy recognition of claim 3, wherein, The scaling the initial foreground image and the initial foreground mask image based on a preset proportion determination rule to obtain a corresponding scaled foreground image and a scaled foreground mask image, includes: randomly generating an initial proportion value based on a preset proportion threshold range of the initial foreground image and the background image of interest; determining a target scaling ratio according to the initial proportion value, the original size ratio of the initial foreground image, and the original size ratio of the background image of interest; scaling the initial foreground image and the initial foreground mask image based on the target scaling ratio to obtain a corresponding scaled foreground image and a scaled foreground mask image.

5. The method of aisle occupancy recognition of claim 1, wherein, The matching template is obtained by performing fusion calculation on the high-dimensional feature map set corresponding to the image category, and the current frame high-dimensional feature map is compared with the matching template based on the comparison result to determine whether to output the image category and the image classification confidence, including: performing weighted fusion calculation on the high-dimensional feature map set corresponding to the image category to obtain a matching template; performing feature comparison between the current frame high-dimensional feature map and the matching template to obtain a corresponding difference value; determine whether the difference value is less than a preset difference threshold, if less than, output the image category and the image classification confidence. 6.The passage occupancy recognition method according to any one of claims 1 to 5, characterized in that, The current frame high-dimensional feature map is updated to the high-dimensional feature map set of the corresponding category based on the buffer fault tolerance mechanism, and the image category and the image classification confidence are output, including: determine whether the state of the buffer set is a direct cache state; if not, determine whether the category corresponding to the current frame high-dimensional feature map is the same as the category corresponding to the previous frame high-dimensional feature map; if the same, update the current frame high-dimensional feature map to the high-dimensional feature map set of the corresponding category, and output the image category and the image classification confidence; If not, the current frame high-dimensional feature map is updated to the buffer set, and the buffer set is marked as the direct cache state.

7. The method of aisle occupancy recognition of claim 6, wherein, After judging whether the state of the buffer set is the direct cache state, the method further includes: If yes, the current frame high-dimensional feature map is directly cached to the buffer set; When the number of high-dimensional feature maps in the buffer set reaches a preset number, the preset number of high-dimensional feature maps in the buffer set are subjected to category analysis to obtain a classification result, and a target classification result with the largest proportion is screened from the classification result; When the proportion corresponding to the target classification result exceeds a preset proportion threshold, a number of high-dimensional feature maps corresponding to the category of the target classification result in the buffer set are updated to a high-dimensional feature map set of the corresponding category, and the buffer set is emptied and marked as an end analysis state.

8. A passage occupancy recognition apparatus characterized by comprising: The method includes: an image of interest extraction module configured to acquire a current frame image and create a mask image with the same size as the current frame image according to a to-be-detected channel region, perform intersection operation on the mask image and the current frame image, and extract a region of interest image corresponding to the to-be-detected channel region from the intersection result; a confidence degree judgment module configured to use a pre-trained binary classification network model to perform channel occupancy identification on the region of interest image to obtain a corresponding image category, an image classification confidence degree, and a current frame high-dimensional feature map, and judge whether the image classification confidence degree is greater than a preset confidence degree threshold; a set updating module configured to, if yes, update the current frame high-dimensional feature map to a high-dimensional feature map set of the corresponding category based on a buffer fault tolerance mechanism, and output the image category and the image classification confidence degree; a feature comparison module configured to, if no, perform fusion calculation on the high-dimensional feature map set corresponding to the image category to obtain a matching template, compare the current frame high-dimensional feature map with the matching template, and determine whether to output the image category and the image classification confidence degree based on a comparison result; The set updating module is specifically configured to, by using the buffer fault tolerance mechanism, cache the current frame high-dimensional feature map to a buffer set for buffering when the image category suddenly changes, cache high-dimensional feature maps with image classification confidence degrees greater than the preset confidence degree threshold in the preset number of frames after the current frame high-dimensional feature map to the buffer set, so that the number of high-dimensional feature maps in the buffer set reaches a preset number, and determine whether to update to the high-dimensional feature map set of the corresponding category after category analysis on the preset number of high-dimensional feature maps stored in the buffer set.

9. An electronic device, comprising: The method includes: a memory configured to save a computer program; a processor configured to execute the computer program to implement the channel occupancy identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is saved, and the computer program is executed by a processor to implement the channel occupancy identification method according to any one of claims 1 to 7.

Images