Intelligent insect monitoring and analyzing system and method

Abstract

The invention provides an insect intelligent monitoring analysis system and method. The method comprises the following steps: acquiring a monitoring video of an insect; extracting a key frame image containing complete characteristics of the insect from the monitoring video, and segmenting the key frame image into a plurality of sub-image blocks; based on the texture gradient distribution of the insects in each sub-image block, determining an identification gain when feature identification is carried out on each sub-image block; clustering each sub-image block according to all the recognition gains to obtain a plurality of clusters, and performing convolution recognition on each cluster based on the feature distribution difference of each cluster to obtain the recognition feature of each cluster; constructing an identification feature vector when the insect is subjected to object identification through all the identification features; and obtaining matching information matched with the identification feature vector in an insect sample library, and further obtaining the identification type of the insect according to the matching information. By adopting the scheme of the invention, accurate identification of the insect matching information in the dynamic monitoring process can be realized.

Description

Technical Field

[0001] This application relates to the technical field of image pattern recognition. More specifically, this application relates to an insect intelligent monitoring and analysis system and method. Background Art

[0002] Image pattern recognition is an important research direction in the fields of artificial intelligence and computer vision. Its core goal is to enable a computer to imitate human cognitive abilities, classify, recognize, and predict input data. The input data can be images, sounds, or other signals. The key to image pattern recognition technology lies in feature extraction and classification. For image data, common features include color, texture, shape, and edges, etc. For speech data, it may involve features such as audio frequency and syntax. Classification is to classify the data into predefined categories according to the extracted features. In recent years, the rise of convolutional neural networks has promoted the development of image pattern recognition technology and achieved remarkable breakthroughs in fields such as image recognition, speech recognition, and natural language processing.

[0003] Certain progress has been made in the field of insect recognition with existing image pattern recognition. Image pattern recognition technology relies on computer vision and machine learning, extracts the appearance features of insects using morphological analysis, and then combines classifiers such as support vector machines or decision trees for recognition. In recent years, deep learning has significantly improved the accuracy of insect recognition and achieved end-to-end recognition. Image pattern recognition has been widely applied to agricultural pest monitoring, ecological research, and biodiversity conservation. However, in the existing technology, it is difficult to capture the subtle feature differences of different species of insects, especially in dynamic monitoring scenarios, and mis-matching is prone to occur. Therefore, how to accurately identify the matching information of insects during the dynamic monitoring process has become a difficult problem faced by the industry. Summary of the Invention

[0004] This application provides an insect intelligent monitoring and analysis system and method, which can accurately identify the matching information of insects during the dynamic monitoring process.

[0005] In a first aspect, this application provides an insect intelligent monitoring and analysis method, including the following steps: Obtain the monitoring video of the insects; Extract key frame images containing the complete features of the insects from the monitoring video, and divide the key frame images into multiple sub-image blocks; Determine the recognition gain when performing feature recognition on each sub-image block based on the texture gradient distribution of the insects in each sub-image block; Cluster each sub-image block according to all the recognition gains to obtain multiple clustering clusters, and perform convolutional recognition on each clustering cluster respectively based on the feature distribution differences of each clustering cluster to obtain the recognition features of each clustering cluster; Construct an identification feature vector for insect object recognition through all the recognition features, obtain the matching information in the insect sample library that matches the identification feature vector, and then obtain the recognition type of the insect based on the matching information.

[0006] In some embodiments, extracting key frame images containing complete insect features from the monitoring video specifically includes: Extract a sequence of consecutive frame images from the monitoring video; Preprocess each consecutive frame image in the sequence of consecutive frame images to obtain a processable image corresponding to each consecutive frame image; Determine the integrity score of the insect area in each processable image; Extract key frame images containing complete insect features from all the processable images based on all the integrity scores.

[0007] In some embodiments, dividing the key frame image into multiple sub-image blocks specifically includes: Set the number of divisions of the key frame image; Evenly divide the key frame image into multiple sub-image blocks according to the number of divisions.

[0008] In some embodiments, determining the recognition gain for feature recognition of each sub-image block based on the texture gradient distribution of the insect in each sub-image block specifically includes: Select a sub-image block as the selected sub-image block; Determine the horizontal gradient component and vertical gradient component of each pixel point in the selected sub-image block; Determine the gradient magnitude of each pixel point based on the horizontal gradient component and vertical gradient component of each pixel point; Determine the texture gradient distribution of the insect in the selected sub-image block based on all the gradient magnitudes; Extract the recognition gain for feature recognition of the selected sub-image block according to the texture gradient distribution; Continue to determine the recognition gain for feature recognition of the remaining sub-image blocks.

[0009] In some embodiments, constructing an identification feature vector for insect object recognition through all the recognition features specifically includes: Determine the contribution degree score of each recognition feature; Extract multiple key features from all the recognition features based on the contribution degree score of each recognition feature; Construct an identification feature vector for insect object recognition from all the key features.

[0010] In some embodiments, obtaining the matching information that matches the identification feature vector in the insect sample library specifically includes: Obtaining the comparison vectors corresponding to all insect samples from the insect sample library; Determining the matching degree between the identification feature vector and each comparison vector, where each insect sample corresponds to a matching degree; Extracting the matching information that matches the identification feature vector from all insect samples according to all the matching degrees.

[0011] In some embodiments, a surveillance video of an insect is obtained through a high-definition camera.

[0012] In a second aspect, the present application provides an insect intelligent surveillance and analysis system, including: An acquisition module, configured to acquire a surveillance video of an insect; A processing module, configured to extract key frame images containing the complete features of the insect from the surveillance video, and segment the key frame images into a plurality of sub-image blocks; The processing module is further configured to determine the recognition gain when performing feature recognition on each sub-image block based on the texture gradient distribution of the insect in each sub-image block; The processing module is further configured to cluster each sub-image block according to all the recognition gains to obtain a plurality of clustering clusters, and perform convolutional recognition on each clustering cluster respectively based on the feature distribution differences of each clustering cluster to obtain the recognition features of each clustering cluster; The processing module is further configured to construct an identification feature vector for object recognition of the insect through all the recognition features; An execution module, configured to obtain the matching information that matches the identification feature vector in the insect sample library, and then obtain the recognition type of the insect according to the matching information.

[0013] In a third aspect, the present application provides a computer device, where the computer device includes a memory and a processor, the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned insect intelligent surveillance and analysis method.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned insect intelligent surveillance and analysis method is implemented.

[0015] The technical solutions provided by the disclosed embodiments of the present application have the following beneficial effects: In the insect intelligent monitoring and analysis system and method provided by the present application, first, a monitoring video of insects is obtained; second, key frame images containing complete insect features are extracted from the monitoring video, and the key frame images are segmented into multiple sub-image blocks; further, based on the texture gradient distribution of insects in each sub-image block, the recognition gain for feature recognition of each sub-image block is determined; then, according to all the recognition gains, each sub-image block is clustered to obtain multiple clustering clusters, and convolutional recognition is performed on each clustering cluster based on the feature distribution difference of each clustering cluster to obtain the recognition features of each clustering cluster; in addition, an identification feature vector for object recognition of insects is constructed through all the recognition features; finally, matching information in the insect sample library that matches the identification feature vector is obtained, and then the recognition type of the insect is obtained based on the matching information.

[0016] It can be seen that the present application can achieve accurate recognition of insect matching information during the dynamic monitoring process; first, key frame images containing complete insect features are extracted from the monitoring video to reflect the complete state of the target object, thereby improving the accuracy of subsequent monitoring and analysis; second, the key frame images are segmented into multiple sub-image blocks, which is beneficial for local feature analysis, ensuring the fineness of the analysis and reducing the computational overhead; further, based on the texture gradient distribution of insects in the sub-image blocks, the recognition gain for feature recognition of the sub-image blocks is determined to determine the importance of the sub-image blocks for extracting the overall insect target features, thereby reasonably allocating computing resources and improving the efficiency of monitoring and analysis; then, multiple clustering clusters of all sub-image blocks are determined, and recognition features are extracted based on the feature distribution differences of the clustering clusters to obtain feature representations for distinguishing and identifying different insect categories, thereby making the subsequent insect classification and matching work more efficient and improving the accuracy of monitoring and analysis; in addition, an identification feature vector for object recognition of insects is constructed through all the recognition features to reflect the visual information of the insects and improve the discrimination of the subtle feature differences between different types of insects, thereby ensuring the accuracy rate when extracting matching information; finally, matching information in the insect sample library that matches the identification feature vector is obtained, and then the recognition type of the insect is obtained based on the matching information to improve the generalization ability of the recognition, thereby ensuring the accuracy of the recognition and providing data support for subsequent related analysis and decision-making; in summary, the technical solution provided by the present application can achieve accurate recognition of insect matching information during the dynamic monitoring process. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is an exemplary flowchart of an insect intelligent monitoring and analysis method according to some embodiments of the present application; Figure 2 is a schematic structural diagram of segmented sub-image blocks according to some embodiments of the present application; Figure 3It is an exemplary flowchart for determining recognition features shown in some embodiments of the present application; Figure 4 It is a schematic structural diagram of an insect intelligent monitoring and analysis system shown in some embodiments of the present application; Figure 5 It is a schematic structural diagram of a computer device for implementing the insect intelligent monitoring and analysis method shown in some embodiments of the present application. Detailed implementation manners

[0018] To better understand the technical solution of the present application, the technical solution of the present application will be described in detail below in conjunction with the accompanying drawings of the specification and specific implementation manners.

[0019] Refer to Figure 1 , this figure is an exemplary flowchart of the insect intelligent monitoring and analysis method shown in some embodiments of the present application. The insect intelligent monitoring and analysis method 100 mainly includes the following steps: In step 101, a monitoring video of the insect is acquired.

[0020] In specific implementation, to acquire a monitoring video of the insect, that is: a monitoring video of the insect is acquired through a camera device for monitoring the activities of the insect. Specifically, in this embodiment, a high-definition camera is selected as the camera device. The camera device can store the captured video data in a server for subsequent calling. In addition, in other embodiments, other shooting devices can also be used to acquire the monitoring video of the insect. For example, a macro camera, an infrared camera, or a camera carried by a drone, etc., which are not limited here.

[0021] It should be noted that the monitoring video of the insect in the present application represents digital video data recording the activities of the insect. By acquiring the monitoring video of the insect, the identification, classification of the insect, and ecological environment monitoring can be performed, and automatic analysis can be carried out in combination with computer vision and image pattern recognition technologies.

[0022] In step 102, key frame images containing the complete features of the insect are extracted from the monitoring video, and the key frame images are segmented into multiple sub-image blocks.

[0023] In some embodiments, the key frame images containing the complete features of the insect can be extracted from the monitoring video in the following manner, that is: Extract a sequence of consecutive frame images from the monitoring video; Preprocess each consecutive frame image in the sequence of consecutive frame images to obtain a processable image corresponding to each consecutive frame image; Determine the integrity score of the insect region in each processable image; Based on all the integrity scores, extract key frame images containing the complete features of the insect from all the processable images.

[0024] In a specific implementation, a continuous frame image sequence is extracted from the surveillance video, that is, the surveillance video is frame decomposed, and redundant frames are removed by using a method based on dynamic time warping. Specifically, Open CV can be used to perform frame decomposition, and then a continuous frame image sequence is extracted from the surveillance video.

[0025] It should be noted that the method based on dynamic time warping in this embodiment is a method for measuring the similarity between adjacent frames. Through this method, frames with large changes can be retained, thereby improving the effectiveness of subsequent insect detection.

[0026] In a specific implementation, each continuous frame image in the continuous frame image sequence is preprocessed to obtain a processable image corresponding to each continuous frame image, that is, each continuous frame image in the continuous frame image sequence is preprocessed, and the preprocessing process includes denoising, grayscale, and histogram equalization, so as to obtain a processable image corresponding to each continuous frame image. In addition, in other embodiments, the preprocessing process may also include other steps, such as adaptive enhancement processing, etc., which are not limited here.

[0027] It should be noted that, in this embodiment, the processable image refers to an image that can be further analyzed. Specifically, in this embodiment, the processable image refers to an image that meets the requirements of the computer vision algorithm in terms of format, color space, resolution, clarity, lighting, etc.

[0028] Specifically, the integrity score of the insect region in each processable image is determined, that is, the insect target in each processable image is detected using a target detection algorithm, the bounding box position of the insect target in each processable image is obtained as the insect region, and the integrity score of the insect region in each processable image is calculated. Specifically, the basis for the insect region integrity score is the completeness of the outer contour of the insect target, the size of the visible area, and the clarity of key parts.

[0029] It should be noted that the integrity score in this embodiment represents a metric for measuring the integrity of insect features. The higher the integrity score, the clearer the insect morphology and the more complete the structure.

[0030] In specific implementation, key frame images containing complete features of insects are extracted from all processable images based on all integrity scores, that is, the highest integrity score is extracted from all integrity scores, and the processable image corresponding to the extracted integrity score is used as the key frame image containing complete features of insects.

[0031] It should be noted that in this application, the key frame image refers to a frame image that can reflect the complete state of the target object. Specifically, in this application, the key frame image is a frame image extracted from a video sequence that can characterize the complete state of the insect. By determining the key frame image, the accuracy of subsequent monitoring and analysis can be improved.

[0032] It should also be noted that in this application, the sub-image block refers to a smaller area segmented from the original image. Specifically, in this application, the sub-image block is a smaller area segmented from the key frame image. By segmenting the sub-image block, it is beneficial for local feature analysis, ensuring the fineness of the analysis, and at the same time reducing the computational overhead. Refer to Figure 2 As shown in the figure, this figure is a schematic structural diagram of the segmented sub-image block according to some embodiments of this application.

[0033] As a preferred embodiment, the key frame image can be segmented into multiple sub-image blocks by the following method, that is: Set the number of segments of the key frame image; According to the number of segments, evenly divide the key frame image into multiple sub-image blocks.

[0034] When specifically implemented, evenly divide the key frame image into multiple sub-image blocks according to the number of segments, that is: obtain the number of segments M*N, evenly divide the key frame image into M rows and N columns of grid areas with equal sizes according to the number of segments, calculate the boundary coordinates of each grid area, and use the slicing tool to segment the key frame image according to the boundary coordinates to obtain multiple sub-image blocks.

[0035] In step 103, determine the recognition gain when performing feature recognition on each sub-image block based on the texture gradient distribution of the insect in each sub-image block.

[0036] In some embodiments, the recognition gain when performing feature recognition on each sub-image block based on the texture gradient distribution of the insect in each sub-image block can be obtained by the following method, that is: Select a sub-image block as the selected sub-image block; Determine the horizontal gradient component and vertical gradient component of each pixel point in the selected sub-image block; Based on the horizontal gradient component and vertical gradient component of each pixel point, determine the gradient amplitude of each pixel point; Based on all the gradient amplitudes, determine the texture gradient distribution of the insect in the selected sub-image block; Extract the recognition gain when performing feature recognition on the selected sub-image block according to the texture gradient distribution; Continue to determine the recognition gain when performing feature recognition on the remaining sub-image blocks.

[0037] In specific implementation, the horizontal gradient component and the vertical gradient component of each pixel in the selected sub-image block are determined, that is: each pixel in the selected sub-image block is calculated through the Sobel gradient operator to obtain the horizontal gradient component and the vertical gradient component of each pixel. In addition, in other embodiments, other methods can also be used to calculate the horizontal gradient component and the vertical gradient component of the pixel, which is not limited here.

[0038] In specific implementation, the gradient magnitude of each pixel is determined based on the horizontal gradient component and the vertical gradient component of each pixel, that is: the square value of the horizontal gradient component of each pixel is added to the square value of the vertical gradient component, and the arithmetic square root of the calculation result is used as the gradient magnitude of each pixel.

[0039] In specific implementation, the texture gradient distribution of the insect in the selected sub-image block is determined based on all the gradient magnitudes, that is: the horizontal gradient component, the vertical gradient component and the gradient magnitude of each pixel are obtained, and the texture gradient distribution of the insect in the selected sub-image block is calculated through the horizontal gradient component, the vertical gradient component and the gradient magnitude of each pixel. The texture gradient distribution includes a gradient histogram, a gradient entropy and a gradient variance. Specifically, the gradient histogram is calculated through the horizontal gradient component and the vertical gradient component, the gradient entropy is calculated through the gradient magnitude, and the gradient variance is calculated through the gradient magnitude. The specific calculation process of the relevant statistical data in the texture gradient distribution is not described in detail here. In addition, in other embodiments, the texture gradient distribution may also include other statistical data, which is not limited here.

[0040] It should be noted that the texture gradient distribution in this application represents the spatial distribution of the pixel gradients in the image. Specifically, the texture gradient distribution in this application refers to the spatial distribution of the gradients of all pixels in the sub-image block. By determining the texture gradient distribution, the texture information of the insect in the image can be effectively described, thereby improving the reliability of the subsequent recognition task.

[0041] Among them, in some embodiments, the recognition gain for feature recognition of the selected sub-image block can be extracted according to the texture gradient distribution in the following manner, that is: Determine the gain weight coefficient of the selected sub-image block; Determine the recognition gain for feature recognition of the selected sub-image block according to the gain weight coefficient and the texture gradient distribution.

[0042] In specific implementation, the gain weight coefficient of the selected sub-image block is determined, that is: the texture gradient distribution of the selected sub-image block is obtained, the gradient histogram, gradient entropy and gradient variance are extracted from the texture gradient distribution, the mean value of the gradient histogram is calculated, and the mean value is weighted and summed with the gradient entropy and gradient variance. The calculation result of the weighted sum is used as the gain weight coefficient of the selected sub-image block. The weight assignment in the specific weighted sum process can be set according to actual application requirements. For example, the weight of the mean value of the gradient histogram is set to 0.4, the weight of the gradient entropy is set to 0.3, and the weight of the gradient variance is set to 0.3.

[0043] It should be noted that in this embodiment, the gain weight coefficient represents an index for measuring the contribution degree of different image regions in the feature extraction process. Specifically, in this embodiment, the gain weight coefficient refers to an index for measuring the contribution degree of different sub-image blocks in the feature extraction process. By determining the gain weight coefficient, the influence of important regions can be enhanced, thereby improving the accuracy of feature extraction.

[0044] In specific implementation, the recognition gain for feature recognition of the selected sub-image block is determined according to the gain weight coefficient and the texture gradient distribution, that is: First, the gain weight coefficients of other sub-image blocks and the selected sub-image block are obtained, and the sum of all gain weight coefficients is calculated; Second, the distribution divergence of the selected sub-image block is calculated according to the texture gradient distribution. Specifically, the Kullback-Leibler method can be used to calculate the distribution divergence, and the specific calculation process is not described here; Finally, the quotient of the distribution divergence and the sum of all gain weight coefficients is calculated, and the quotient result is multiplied by the gain weight coefficient of the selected sub-image block. The multiplication result is used as the recognition gain for feature recognition of the selected sub-image block.

[0045] It should be noted that in this application, the recognition gain represents the degree of improvement in the ability of feature information in the image to distinguish categories in pattern recognition. Specifically, in this application, the recognition gain refers to a quantitative index for evaluating the discrimination ability of insect features in sub-image blocks. The larger the recognition gain, the stronger the discrimination ability of insect features in the sub-image block, and the smaller the recognition gain, the weaker the discrimination ability of insect features in the sub-image block. By determining the recognition gain, the computing resources can be reasonably allocated, thereby improving the efficiency of monitoring and analysis.

[0046] In step 104, each sub-image block is clustered according to all the recognition gains to obtain a plurality of clustering clusters, and convolution recognition is performed on each clustering cluster respectively based on the feature distribution differences of each clustering cluster to obtain the recognition features of each clustering cluster.

[0047] In some embodiments, clustering each sub-image block according to all the recognition gains to obtain a plurality of clustering clusters can be performed in the following manner, that is: Determine the normalization parameter corresponding to each recognition gain; Divide all sub-image blocks into multiple clustering clusters according to the normalization parameter corresponding to each sub-image block.

[0048] In specific implementation, determine the normalization parameter corresponding to each recognition gain, that is: use the Z-score normalization method to normalize each recognition gain to obtain the normalization parameter corresponding to each recognition gain. In addition, in other embodiments, other methods can also be used for normalization, such as the Min-Max method, etc., which are not limited here.

[0049] In specific implementation, divide all sub-image blocks into multiple clustering clusters according to the normalization parameter corresponding to each sub-image block, that is: according to the normalization parameter corresponding to each sub-image block, combine the K-means clustering method to cluster all sub-image blocks, and then divide all sub-image blocks into multiple clustering clusters. In addition, in other embodiments, other methods can also be used for clustering, such as Mean-Shift, etc., which are not limited here.

[0050] It should be noted that in this application, a clustering cluster represents a group of data with similar characteristics. Specifically, in this application, a clustering cluster refers to a group of sub-image blocks with similar recognition gains. Each clustering cluster contains multiple sub-image blocks with similar recognition gains. Determining the clustering cluster is beneficial to the monitoring and analysis of insects.

[0051] In some embodiments, refer to Figure 3 As shown, this figure is an exemplary flowchart for determining recognition features according to some embodiments of this application. In this embodiment, convolution recognition is performed on each clustering cluster respectively based on the feature distribution differences of each clustering cluster, and the recognition features of each clustering cluster can be implemented by the following steps: First, in step 1041, determine the feature distribution differences of each clustering cluster; Then, in step 1042, allocate adaptive convolution kernel parameters for each clustering cluster based on the feature distribution differences corresponding to each clustering cluster; Furthermore, in step 1043, for all sub-image blocks in each clustering cluster, extract the depth features of each sub-image block according to the convolution kernel parameters corresponding to the clustering cluster; Finally, in step 1044, determine the recognition features of the clustering cluster based on all the depth features, and then obtain the recognition features of each clustering cluster.

[0052] Among them, in some embodiments, the feature distribution differences of each clustering cluster can be determined in the following way, that is: Select a clustering cluster as the selected clustering cluster; Determine the feature distribution coefficient of the selected clustering cluster; Extract the feature distribution difference of the selected clustering cluster based on the feature distribution coefficient; Continue to determine the feature distribution difference of each remaining clustering cluster.

[0053] In specific implementation, determine the feature distribution coefficient of the selected clustering cluster, that is: select a sub-image block from the selected clustering cluster, detect the effective area containing the insect feature part in the sub-image block through a morphological-based detection algorithm, and calculate the pixel area of the effective area through OpenCV, so as to obtain the pixel area corresponding to each sub-image block in the selected clustering cluster, and take the mean value of all pixel areas as the feature distribution coefficient of the selected clustering cluster.

[0054] It should be noted that in this embodiment, the feature distribution coefficient represents a quantitative index for describing the data feature distribution. Specifically, in this embodiment, the feature distribution coefficient refers to the index of the area size of the insect feature area of the clustering cluster during the monitoring and analysis process.

[0055] In specific implementation, extract the feature distribution difference of the selected clustering cluster based on the feature distribution coefficient, that is: obtain the feature distribution coefficients of the selected clustering cluster and other clustering clusters, calculate the sum of all feature distribution coefficients, and take the quotient of the feature distribution coefficient of the selected clustering cluster and the sum as the feature distribution difference of the selected clustering cluster.

[0056] It should be noted that in this application, the feature distribution difference represents the distribution form difference of different clustering clusters in the feature space. Specifically, in this application, the feature distribution difference refers to the distribution difference of the effective feature parts of the sub-image blocks in different clustering clusters in terms of area. By determining the feature distribution difference, similarity discrimination can be made for different clustering clusters.

[0057] In specific implementation, allocate adaptive convolution kernel parameters for each clustering cluster based on the feature distribution difference corresponding to each clustering cluster, that is: according to the numerical value of the feature distribution difference corresponding to each clustering cluster, dynamically adjust the scale size of the convolution kernel through an adaptive convolution method, so as to allocate adaptive convolution kernel parameters for each clustering cluster. In addition, in other embodiments, other methods can also be used to allocate convolution kernel parameters, for example, deformable convolution method, etc., which are not limited here.

[0058] In specific implementation, extract the deep features of each sub-image block according to the convolution kernel parameters corresponding to the clustering cluster, that is: call the deep convolutional neural network model, and perform convolution on each sub-image block in the clustering cluster through the deep convolutional neural network model combined with the convolution kernel parameters corresponding to the clustering cluster. The specific convolution implementation process is not described here again, so as to obtain the deep features of each sub-image block in the clustering cluster.

[0059] It should be noted that in this embodiment, the depth feature representation refers to the feature information calculated by a deep neural network in a certain image region. Specifically, in this embodiment, the depth feature refers to the feature information of a sub-image block. By extracting the depth feature, the data can be made more discriminative and hierarchical.

[0060] When specifically implemented, the recognition feature of the clustering cluster is determined based on all the depth features, that is: the feature fusion of all the depth features is performed by the weighted feature projection method, and the fused feature is used as the recognition feature of the clustering cluster. Specifically, first, calculate the Fisher Score index of each depth feature as the weight of each depth feature; second, multiply each depth feature by its corresponding weight to obtain the weighted feature; then, project all the weighted features through a non-linear mapping to reduce the data dimension and enhance the discriminative ability of the feature; finally, splice all the projected features column by column to obtain the recognition feature of the clustering cluster. In addition, in other embodiments, other methods can also be used to determine the recognition feature, for example, the adaptive feature fusion method, etc., which are not limited here.

[0061] It should be noted that in this application, the recognition feature representation is used to distinguish and identify different target categories. Specifically, in this application, the recognition feature refers to the feature representation used to distinguish and identify different insect categories. The recognition feature has strong discriminative ability, so that the subsequent insect classification, matching or retrieval work is more efficient, and the accuracy of monitoring and analysis is improved.

[0062] In step 105, an identification feature vector for object recognition of insects is constructed through all the recognition features.

[0063] In some embodiments, the identification feature vector for object recognition of insects can be constructed through all the recognition features in the following manner, that is: Determine the contribution degree score of each recognition feature; Extract multiple key features from all the recognition features based on the contribution degree score of each recognition feature; Construct an identification feature vector for object recognition of insects from all the key features.

[0064] When specifically implemented, determine the contribution degree score of each recognition feature, that is: evaluate each recognition feature through the chi-square test respectively, and use the obtained chi-square statistic as the contribution degree score of each recognition feature. In addition, in other embodiments, other methods can also be used for evaluation, for example, the F-test, etc., which are not limited here.

[0065] It should be noted that in this embodiment, the contribution score is an important indicator for measuring the role of a single feature in target monitoring. Specifically, in this embodiment, the contribution score refers to the strength of the correlation between the recognition feature and the insect. The higher the contribution score, the stronger the correlation between the recognition feature and the insect; the lower the contribution score, the weaker the correlation between the recognition feature and the insect. Determining the contribution score is beneficial for feature screening, thereby optimizing the monitoring efficiency.

[0066] In specific implementation, multiple key features are extracted from all the recognition features based on the contribution scores of each recognition feature, that is: a judgment threshold is set, all the contribution scores are compared with the judgment threshold, and the recognition features with all contribution scores greater than the judgment threshold are extracted, and the extracted recognition features are used as key features.

[0067] It should be noted that in this embodiment, the key feature represents a feature containing effective recognition information in the monitoring task. Extracting the key feature can improve the reliability of the monitoring.

[0068] In specific implementation, an identification feature vector for object recognition of insects is constructed from all the key features, that is: the contribution score corresponding to each key feature is obtained, all the key features are sorted in descending order based on the contribution scores corresponding to each key feature to obtain a sorting sequence, and all the key features are vector-combined according to the sorting order of the sorting sequence to obtain an identification feature vector for object recognition of insects.

[0069] It should be noted that in this application, the identification feature vector represents a high-dimensional numerical vector for describing the target. Specifically, in this application, the identification feature vector is a high-dimensional numerical vector for describing the insect target. The identification feature vector contains multiple discriminative features. Determining the identification feature vector can reflect the visual information of the insect, thereby increasing the effectiveness of the recognition.

[0070] In step 106, the matching information matching the identification feature vector in the insect sample library is obtained, and then the recognition type of the insect is obtained based on the matching information.

[0071] In some embodiments, the matching information matching the identification feature vector in the insect sample library can be obtained in the following manner, that is: The comparison vectors corresponding to all insect samples are obtained from the insect sample library; The matching degree between the identification feature vector and each comparison vector is determined, where each insect sample corresponds to a matching degree; The matching information matching the identification feature vector is extracted from all the insect samples based on all the matching degrees.

[0072] It should be noted that in this embodiment, the comparison vector represents the only vector describing the characteristics of the insect, and the comparison vector can reflect information such as the type, texture, shape, and color of the insect.

[0073] Specifically, when implementing, determine the matching degree between the identification feature vector and each comparison vector, that is: calculate the similarity metric value between the identification feature vector and each comparison vector based on the similarity matching rule. Specifically, calculate the similarity metric value between the identification feature vector and each comparison vector through the cosine similarity calculation method, and use the calculation result as the matching degree between the identification feature vector and each comparison vector. In addition, in other embodiments, other similarity matching rules can also be used to determine the matching degree. For example, the Euclidean distance calculation method, the Manhattan distance calculation method, etc. are not limited here.

[0074] It should be noted that in this embodiment, the similarity matching rule represents an algorithm for measuring and comparing the similarity between two objects. This rule evaluates the similarity degree of two objects in the feature space based on a specific metric standard.

[0075] Specifically, when implementing, extract the matching information that matches the identification feature vector from all insect samples according to all the matching degrees, that is: obtain all insect samples and set a similarity threshold, compare all the matching degrees with the similarity threshold, extract the insect samples whose all matching degrees are greater than the similarity threshold, and construct the matching information that matches the identification feature vector according to the extracted insect samples.

[0076] It should be noted that in this application, the matching information represents the detailed information of the samples similar to the input data. Specifically, the matching information in this application refers to the detailed information of the samples that match the identification feature vector of the insect. The matching information includes the type, sample number, and external attributes of the insects corresponding to all the extracted insect samples. Determining the matching information is beneficial to more accurately complete the object recognition task.

[0077] Specifically, when implementing, obtain the recognition type of the insect according to the matching information, that is: count the number of occurrences of each insect type in the matching information, extract the insect type with the most occurrences, and use the extracted insect type as the recognition type of the insect in the intelligent monitoring and analysis task, which improves the generalization ability of the recognition, thus ensuring the accuracy of the recognition and providing data support for subsequent relevant analysis and decision-making.

[0078] In addition, on the other hand of this application, in some embodiments, this application provides an insect intelligent monitoring and analysis system. Refer to Figure 4, This figure is a schematic structural diagram of an insect intelligent monitoring and analysis system according to some embodiments of the present application. The insect intelligent monitoring and analysis system 200 includes: an acquisition module 201, a processing module 202, and an execution module 203, which are described as follows: The acquisition module 201. In the present application, the acquisition module 201 is mainly used to acquire monitoring videos of insects; The processing module 202. In the present application, the processing module 202 is mainly used to extract key frame images containing the complete features of insects from the monitoring videos, and divide the key frame images into multiple sub-image blocks; The processing module 202 is further used to determine the recognition gain when performing feature recognition on each sub-image block based on the texture gradient distribution of insects in each sub-image block; Then, the processing module 202 is further used to cluster each sub-image block according to all the recognition gains to obtain multiple clustering clusters, and perform convolutional recognition on each clustering cluster based on the feature distribution differences of each clustering cluster to obtain the recognition features of each clustering cluster; In addition, the processing module 202 is further used to construct an identification feature vector for object recognition of insects through all the recognition features; The execution module 203. In the present application, the execution module 203 is mainly used to obtain matching information in the insect sample library that matches the identification feature vector, and then obtain the recognition type of the insect based on the matching information.

[0079] In addition, the present application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to obtain the code and execute the above-mentioned insect intelligent monitoring and analysis method.

[0080] In some embodiments, refer to Figure 5 , This figure is a schematic structural diagram of a computer device for implementing the insect intelligent monitoring and analysis method according to some embodiments of the present application. The insect intelligent monitoring and analysis method in the above embodiments can be implemented by Figure 5 the computer device shown. The computer device 300 includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0081] The processor 301 can be a general-purpose central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more for controlling the execution of the insect intelligent monitoring and analysis method in the present application.

[0082] The communication bus 302 can be used to transfer information between the above components.

[0083] The memory 303 can be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or it can also be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 303 can exist independently and be connected to the processor 301 through the communication bus 302. The memory 303 can also be integrated with the processor 301.

[0084] Among them, the memory 303 is used to store the program code for executing the solution of this application, and is controlled by the processor 301 to execute. The processor 301 is used to execute the program code stored in the memory 303. The program code can include one or more software modules. The determination of the insect intelligent monitoring and analysis method in the above embodiments can be implemented by one or more software modules in the program code in the processor 301 and the memory 303.

[0085] The communication interface 304, using any device such as a transceiver, is used to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0086] In a specific implementation, as an embodiment, the computer device can include multiple processors, and each of these processors can be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, the processor can refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions).

[0087] The computer device described above can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a laptop computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of the present application do not limit the type of the computer device.

[0088] In addition, the present application also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above-mentioned insect intelligent monitoring and analysis method is implemented.

[0089] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0090] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. An insect intelligent monitoring and analysis method, characterized in that, It includes the following steps: Obtain the surveillance video of the insect; Extract the key frame images containing the complete features of the insect from the surveillance video, and segment the key frame images into multiple sub-image blocks; Determine the recognition gain when performing feature recognition on each sub-image block based on the texture gradient distribution of the insect in each sub-image block; Cluster each sub-image block according to all the recognition gains to obtain multiple clustering clusters, and perform convolutional recognition on each clustering cluster respectively based on the feature distribution differences of each clustering cluster to obtain the recognition features of each clustering cluster; Construct an identification feature vector for object recognition of the insect through all the recognition features; Obtain the matching information in the insect sample library that matches the identification feature vector, and then obtain the recognition type of the insect based on the matching information.

2. The method according to claim 1, characterized in that, Specifically, extracting the key frame images containing the complete features of the insect from the surveillance video includes: Extract a sequence of consecutive frame images from the surveillance video; Preprocess each consecutive frame image in the sequence of consecutive frame images to obtain a processable image corresponding to each consecutive frame image; Determine the integrity score of the insect area in each processable image; Extract the key frame images containing the complete features of the insect from all the processable images based on all the integrity scores.

3. The method according to claim 1, characterized in that, Specifically, segmenting the key frame images into multiple sub-image blocks includes: Set the number of segments of the key frame image; Evenly divide the key frame image into multiple sub-image blocks according to the number of segments.

4. The method according to claim 1, characterized in that Specifically, determining the recognition gain when performing feature recognition on each sub-image block based on the texture gradient distribution of the insect in each sub-image block includes: Select a sub-image block as the selected sub-image block; Determine the horizontal gradient component and vertical gradient component of each pixel point in the selected sub-image block; Determine the gradient amplitude of each pixel point based on the horizontal gradient component and vertical gradient component of each pixel point; Determine the texture gradient distribution of the insect in the selected sub-image block based on all the gradient amplitudes; Extract the recognition gain when performing feature recognition on the selected sub-image block according to the texture gradient distribution; Continue to determine the recognition gain when performing feature recognition on the remaining sub-image blocks.

5. The method according to claim 1, characterized in that, Specifically, constructing an identification feature vector for object recognition of the insect through all the recognition features includes: Determine the contribution degree score of each recognition feature; Extract multiple key features from all the recognition features based on the contribution degree score of each recognition feature; Construct an identification feature vector for object recognition of the insect from all the key features.

6. The method according to claim 1, wherein Specifically, obtaining the matching information in the insect sample library that matches the identification feature vector includes: Obtain the comparison vectors corresponding to all insect samples from the insect sample library; Determine the matching degree between the identification feature vector and each comparison vector, where each insect sample corresponds to a matching degree; Extract the matching information that matches the identification feature vector from all the insect samples based on all the matching degrees.

7. The method according to claim 1, characterized in that, Obtain the surveillance video of the insect through a high-definition camera.

8. An insect intelligent monitoring and analysis system, characterized in that, It includes: An acquisition module for acquiring the surveillance video of the insect; A processing module, configured to extract key frame images containing complete insect features from the monitoring video, and segment the key frame images into a plurality of sub-image blocks; The processing module is further configured to determine an identification gain for feature identification of each sub-image block based on the texture gradient distribution of the insects in each sub-image block; The processing module is further configured to cluster each sub-image block according to all the identification gains to obtain a plurality of clustering clusters, and perform convolutional identification on each clustering cluster respectively based on the feature distribution differences of each clustering cluster to obtain the identification features of each clustering cluster; The processing module is further configured to construct an identification feature vector for object identification of insects through all the identification features; An execution module, configured to obtain matching information in an insect sample library that matches the identification feature vector, and further obtain the identification type of the insect according to the matching information.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory stores code, and the processor is configured to obtain the code and execute the insect intelligent monitoring and analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the insect intelligent monitoring and analysis method according to any one of claims 1 to 7.

Images