31 results about "Texture extraction" patented technology

The application discloses a multi-focus microscopic image fusion method, system and program based on depth map correction. The method comprises the following steps in sequence: image sequence acquisition, texture extraction, definition evaluation, depth map generation, confidence map generation, depth correction and color image reconstruction. According to the definition curve, the depth map and the confidence map are determined, the unstable points are determined by using the confidence map, the depth data of the non-texture and weak-texture areas are corrected by HSL color space guided filtering, and thus the phenomenon that the depth map is wrong in focus plane judgment due to the non-texture and weak-texture areas of the image source is avoided.

The invention relates to a small target detection method and system based on adaptive texture perception and frequency domain-cross-layer collaborative optimization, and the method comprises the steps: providing a cross-scale texture guide module, independently capturing the spatial distribution characteristics of heterogeneous features through a multi-scale texture extraction unit, and constructing scene-level texture prior through an additive fusion mechanism; a frequency domain detail enhancement module is provided, features are mapped to a complex frequency spectrum space through real number fast Fourier transform, dynamic reweighting is carried out on medium-high frequency components by using an adaptive weight vector, and target contour information is explicitly enhanced in combination with inverse transform and a residual connection mechanism; a cross-layer feature stabilization module is designed, context clues of different granularities are captured through lightweight multi-scale receptive field extension, and a dynamic channel gating modulation mechanism is introduced to perform active intervention on a feature evolution path. Compared with the prior art, the method has the advantages of characteristic enhancement, stable positioning, balanced detection precision and operation efficiency and the like.

The application provides a polarization and visible light image fusion texture extraction method and device. A visible light reflection component is first extracted to screen a reliable linear polarization degree image texture and generate a preliminary fusion texture image. Then, a deep learning network model is trained by taking the preliminary fusion texture as a supervised true value, combining a reasonable data enhancement method and a scientific training strategy, so as to eliminate residual block effects and noises in artificial textures. Finally, a fusion texture with high information quantity, high precision, low noise and low block effect is output. In this way, the complementary characteristics of the polarization and visible light images are fully utilized, the texture extraction result is clear and free of artifacts, and the method is suitable for image fusion, edge detection and other real complex scene image detail enhancement and quality optimization.

The invention relates to the technical field of image recognition, in particular to a catheter positioning method and system based on image recognition, and the method comprises the steps: extracting multi-direction gray changes from a catheter image, generating main direction angle mapping, comparing adjacent pixel directions to form an edge continuous set, recognizing the trend, matching direction textures, and extracting an extension path. And constructing a connected node chain by screening directions and gray features, and connecting node coordinates to generate a catheter positioning path set. According to the method, a main direction angle distribution diagram is constructed through multi-direction gray scale changes, the edge trend with extension features is extracted through direction continuity and angle stability, linear track classification is completed through direction texture comparison, a coherent node link is obtained through synchronous constraint of direction and gray scale trend, and the linear track classification is completed. A conduit track set with direction correlation and node connection is formed through spatial serialization, path derivation continuity and spatial structure integrity are kept in an area with complex texture or frequent structure change, and the phenomena of trend drift and path jump are reduced.

This invention discloses a crop disease diagnosis method that integrates spatial multi-scale perception with environmental decoupling, belonging to the field of agricultural remote sensing image processing technology. It utilizes large-kernel deep convolution to perform wide-area spatial perception on remote sensing images, generating a spatial weight map encoding global environmental background information. Subsequently, the spatial weight map is reshaped into a dynamic convolution kernel, driving small-kernel group convolution to perform context-adaptive fine-grained disease texture extraction within local neighborhoods, achieving cross-scale feature collaborative modeling. Deep features are decomposed into disease features and environmental features. Adversarial training using gradient inversion layers forces disease features to be free from environmental interference, and mutual information minimization and counterfactual reinforcement learning eliminate environmental spurious correlations, ultimately obtaining robust disease diagnosis results to environmental changes. This invention effectively solves the problem of insufficient generalization performance of existing technologies under varying environmental conditions and has significant application value in multi-temporal and multi-plot agricultural remote sensing monitoring scenarios.

The present application belongs to the technical field of computer vision and intelligent agriculture, and particularly relates to a remote sensing image cultivated land fine segmentation method and system fusing global-local feature perception. The method comprises: acquiring cultivated land remote sensing image data and performing preprocessing; constructing a feature encoder based on a ResNet architecture to perform feature extraction on the preprocessed cultivated land remote sensing image data to obtain shallow features and deep features; constructing a GLC-Mamba bottleneck module to perform local texture extraction and global long-distance dependence modeling on the deep features, and obtaining fused features through a gated adaptive fusion mechanism; and constructing a U-Net decoder to up-sample the fused features and fuse the shallow features through a skip connection to obtain a pixel-level cultivated land segmentation result map. The present application solves the problems of strong spatio-temporal heterogeneity and easy-to-fuzz boundary in cultivated land segmentation, maintains a low computing cost while ensuring high precision, and is suitable for large-scale agricultural condition monitoring and land resource management.

This invention discloses an intelligent fire detection method based on artificial intelligence video analysis, belonging to the field of artificial intelligence video analysis technology. This method targets multiple video streams deployed on an edge computing gateway. It utilizes an asynchronous decoupled buffer queue to bind video frames and timestamps into frame units to be processed. An incremental sequence identifier is assigned to each frame unit to be processed via a hardware resource scheduler, and the frame units are then distributed to a first processing pipeline and a second processing pipeline based on the sequence identifier. In the first processing pipeline, background removal and dynamic texture extraction are performed to generate a thermal activation map. In the second processing pipeline, frequency domain transformation and multi-scale local binary mode decomposition are performed to generate a flame frequency domain texture feature map. The thermal activation map and the flame frequency domain texture feature map are then pixel-level weighted fusion at the same time reference to generate a fusion confidence tensor. Finally, non-maximum suppression and spatial connectivity analysis are performed to output the bounding box of the flame target and the initial fire intensity level.

The invention relates to the field of cargo identification, and discloses a pallet cargo identification method and device and a storage medium. The method comprises the following steps: collecting a depth point cloud and a color image; carrying out cargo extraction processing on the depth point cloud to obtain a cargo point cloud cluster; performing projection cutting processing on the color image to obtain a cargo image block, and cutting a point cloud cluster; performing geometric feature extraction on the cutting point cloud cluster to generate query features; performing texture extraction on the cargo image blocks to generate texture features; performing dot product normalization processing on the query features and the texture features to generate similarity features; performing feature fusion processing on the similarity features and the cargo image blocks to generate cargo fusion features; and based on a preset classifier and a preset regression device, carrying out classification identification processing on the cargo fusion features, and generating cargo type and pose data. In the embodiment of the invention, the cargo identification precision is improved, the influence of illumination on the identification result is overcome, and the data cost and deployment time of new SKU online are reduced.

The application discloses an image texture extraction method, device and computer readable storage medium, the method comprising: converting the color value of each pixel of a target image into a different base to obtain a digital array of each pixel; wherein the base of the color value of the converted pixel is lower than the base of the color value of the pixel before conversion; according to the number of bits of the digital array, the target image is decomposed into several layers; for each layer, a window mask is used to extract texture information to obtain a sub-texture map of each layer; and all the sub-texture maps are superimposed with weights to synthesize a total texture map of the target image. Compared with traditional edge detection technology, the application can greatly improve the texture information extraction rate and texture information extraction accuracy of the target image, and provide more complete texture and edge information for downstream image information processing procedures.

The present application belongs to the technical field of image processing, and particularly relates to a quantum heuristic progressive focusing plant cell microtubule image segmentation method and system. The method of the present application takes quantum heuristic progressive focusing Transformer as the backbone network, takes the feature texture extraction module as the bypass network, and loads the Hamilton growth layer at the output end of the segmentation head to form a plant cell microtubule segmentation network; each Transformer layer of the backbone network comprises, in sequence, a progressive attention layer, a full connection layer, a normalization layer, a global state cognition module, a Bell non-local module and a quantum interference gate. The multi-quantum feature coupling mechanism composed of the global state cognition module, the Bell non-local module and the quantum interference gate effectively solves the segmentation fracture and blur problem of the plant cell microtubule image under the condition of few samples due to low signal-to-noise ratio and complex structure, and improves the accuracy, topological integrity and robustness of segmentation.

This invention relates to the field of image processing technology and proposes a method and system for detecting Ulva prolifera in remote sensing images based on deep learning. The method includes the following steps: for the acquired remote sensing image to be detected, edge gradient extraction and small target enhancement processing are performed to obtain a first feature map after edge enhancement and small target feature enhancement; the first feature map is transmitted to a U-shaped backbone network composed of cascaded multi-level Ep-VSS block modules for multi-level feature extraction; and the detection result of the Ulva prolifera remote sensing image is obtained based on the feature map output by the last-level Ep-VSS block module. This invention achieves accurate boundary segmentation and long-range dependency modeling through edge enhancement, sensitive small target detection, multi-scale texture extraction, and spatial context modeling, thereby improving the accuracy, real-time performance, and reliability of Ulva prolifera remote sensing monitoring.

The invention is applicable to the technical field of color sorters, and provides a color sorter impurity distinguishing method based on texture feature extraction, which comprises the following steps: S1, image acquisition: acquiring an original image of a to-be-detected material through an industrial camera of a color sorter; according to the color sorter impurity distinguishing method based on texture feature extraction, two texture extraction algorithms of gray level co-occurrence matrix and Gabor wavelet transform are fused, texture features are extracted from two dimensions of spatial distribution and direction scale, and compared with a single texture extraction method, the texture difference between materials and impurities can be reflected more comprehensively, so that the color sorter impurity distinguishing method based on texture feature extraction is more accurate. The method has higher distinguishing precision especially for impurities with similar colors and different textures, dimension reduction processing is carried out on initial texture features through a principal component analysis algorithm, redundant features are effectively eliminated, the calculation amount of a classification model is reduced, the sorting efficiency of a color sorter is improved, and the requirement for high-speed material sorting is met. And meanwhile, a support vector machine classification model is adopted, so that the model generalization ability and the classification stability are improved.

The invention discloses a casting casting head segmentation method based on a texture decoupling and structure perception segmentation network. The method comprises the steps that a casting image is collected and preprocessed; constructing a texture decoupling and structure perception segmentation network, wherein the texture decoupling and structure perception segmentation network comprises a background texture extraction model and a casting head segmentation model; casting background texture features of the casting head segmentation data set are extracted based on the trained background texture extraction model; training a casting head segmentation model based on the casting background texture features and the casting head segmentation data set to obtain a trained casting head segmentation model; the casting head segmentation model comprises a global feature extraction model, a segmentation decoder and an output module; and carrying out actual casting head identification based on the trained casting head segmentation model to obtain a casting head segmentation result. The model is deployed to an industrial site, casting images captured by a camera can be processed in real time, and accurate casting head position and form information is output, so that an automatic robot is guided to carry out accurate casting head removal operation.

The invention relates to the technical field of image forgery detection, and discloses a diffusion model image restoration forgery-oriented multi-view fusion and regional decoupling detection system, which comprises a forgery trace extraction module and a tampered region positioning fusion module, the forgery trace extraction module comprises a noise residual extractor, a high-frequency texture extractor, a cross-view fusion device and a multi-stage contrast learning device. According to the system, through a multi-view feature fusion mechanism and a region decoupling strategy, the detection problem caused by visual consistency, edge smoothness and noise distribution homogeneity of a diffusion model repair image is effectively solved; the innovative multi-level contrast learning framework can forcibly separate the feature representation of the counterfeit and real areas, and significantly improve the ability to capture weak counterfeit traces; the dynamic interactive fusion module realizes collaborative optimization of main body and edge features through a multi-scale channel attention mechanism; the DMIL-Net can rapidly and accurately mark the diffusion repair area in the image, and the manual checking burden is relieved.

The application relates to the field of intelligent decision-making, and discloses a face living body detection method, device and equipment in a motion state scene and a medium. The method comprises the following steps: collecting a face image in a motion state scene, performing face detection on the face image to obtain a detected face, performing format standardization on the detected face to obtain a standardized face; performing texture coding on the standardized face to obtain coded texture of the standardized face, extracting texture information, calculating pixel change information, and constructing three-dimensional structure information of the standardized face; respectively performing feature extraction on the texture information, the pixel change information and the three-dimensional structure information to obtain texture features, pixel change features and three-dimensional structure features; performing feature fusion on the texture features, the pixel change features and the three-dimensional structure features to obtain fused features; and calculating a living body detection score of the standardized face to determine a face living body detection result of the face image. The application can improve the comprehensiveness of face living body detection in a motion state scene.

The present application relates to machine vision and image processing technical field, specifically to a kind of workpiece defect feature point's visual image enhancement processing method, including original gradient field generation step: the information of workpiece surface illumination reflection is converted into two-dimensional vector field;Local texture extraction step: introduce structure tensor field and obtain local texture direction field by eigenvalue decomposition;Background flow field construction step: carry out smoothing regularization processing to generate background flow field removed local disturbance;Signal geometry separation step: original gradient field is projected and decomposed into along-texture and inverse-texture gradient component;Gain reconstruction enhancement step: component is handled and reconstructed using differential gain control, and output enhanced image;The present application utilizes the adaptive characteristics of structure tensor field, and realizes the accurate stripping of weak defect signal under strong texture interference.

The invention discloses a grassland mousehole detection method based on image analysis, and relates to the technical field of intelligent image recognition, and the method comprises the following steps: constructing a shallow convolution texture extraction structure, carrying out the continuous texture separation processing of an input grassland image, extracting weak texture information, filtering a brightness jump component, and generating a stable basic texture image; the input module is used for providing input data for subsequent feature fusion; and inputting the basic texture image into a bidirectional feature fusion structure, executing weighted fusion in a plurality of scale channels, weakening highlight interference caused by wet region reflection step by step, and outputting a fusion feature image of which the brightness disturbance is suppressed as input of subsequent feature regulation and control. Wet area reflection is weakened through shallow convolution texture extraction and bidirectional feature fusion, and the texture is kept stable; by combining dynamic weight regulation and difficult sample enhanced loss enhanced mouse hole characteristics, reflection interference is inhibited, stable detection and accurate positioning under illumination variation are realized, and the detection precision and result consistency are effectively improved.

The invention belongs to the technical field of image processing, and particularly relates to a quantum heuristic progressive focusing plant cell microtubule image segmentation method and system. According to the method, a quantum heuristic progressive focusing Transform serves as a backbone network, a feature texture extraction module serves as a bypass network, a Hamiltonian growth layer is loaded at the output end of a segmentation head, and a plant cell microtubule segmentation network is formed; each Transform layer of the backbone network sequentially comprises a progressive attention layer, a full connection layer, a normalization layer, a global state cognition module, a Bell non-localization module and a quantum interference gate. Through a multi-quantum feature coupling mechanism composed of the global state cognition module, the Bell non-localization module and the quantum interference gate, the problem of segmentation fracture and fuzziness caused by low signal-to-noise ratio and complex structure of the plant cell microtubule image under the condition of few samples is effectively solved, and the segmentation accuracy, topological integrity and robustness are improved.

The application provides a method and system for image enhancement by speckle noise filtering of an ultrasonic image, and the method comprises the following steps: carrying out brightness equalization processing and smoothing filtering on a noisy ultrasonic image; eliminating residual textures in the filtered ultrasonic image to extract components of a complete noise mask; and outputting an image after restoring a real noise-free image. The application equivalently regards an ultrasonic imaging signal as an additive signal of a real ultrasonic imaging signal and a stray noise signal of ultrasonic in a tissue; regards speckle noise and other noises as a whole for processing, and equivalently regards it as a noise mask; calculates the noise mask, and obtains a noise-free real ultrasonic image through simple operation; and the ultrasonic image obtained through the processing method of the application not only removes speckle noise in the ultrasonic image, but also well retains image texture details, and the image contrast is stronger.

The application discloses a method for improving the silk printing precision of a display module by using an image enhancement algorithm and relates to the technical field of image enhancement. The method first collects the visible light and near-infrared waveband images of a display module substrate, suppresses the background noise of the substrate and enhances the weak texture of the silk printing edge through multi-spectral differential operation, and generates a fusion enhancement image; then a pixel-level texture generation diffusion model is adopted to complete the silk printing edge texture in combination with a boundary constraint condition to generate a high-resolution silk printing alignment reference image; finally, the alignment deviation correction amount is obtained through pixel-level hierarchical feature matching, and the matching result is fed back to optimize the multi-spectral waveband weight. The application synchronously realizes the collaborative optimization of the silk printing geometric alignment precision and the photoelectric performance of the display module, solves the problems of low imaging contrast of the transparent substrate, difficulty in weak texture extraction and display defects caused by silk printing deviation, improves the silk printing precision, alignment robustness and system real-time performance, and is suitable for the silk printing production scene of a high-precision display module.

The embodiment of the application provides a kind of data visualization method and related equipment, belong to data processing technical field.The method comprises: extracting the scalar data of original multidimensional data, color band texture is generated according to scalar data;Extract the vector data of original multidimensional data, generate latitude texture and longitude texture based on vector data;Current particle position is obtained, and speed vector is calculated based on current particle position, latitude texture and longitude texture, and speed vector is rendered into speed texture;Next particle position texture is calculated and rendered based on current particle position and speed texture;Post-processing position texture is generated and rendered based on next particle position texture and speed texture;Last particle position is obtained, and screen coordinates are calculated based on last particle position, current particle position, post-processing position texture and color band texture, screen coordinates are rendered and handled, and frame buffer is obtained.The embodiment of the application can realize the fast rendering of large amount of NC data in web end.

The embodiment of the invention provides a data visualization method and related equipment, and belongs to the technical field of data processing. The method comprises the following steps: extracting scalar data of original multi-dimensional data, and generating a color tape texture according to the scalar data; extracting vector data of the original multi-dimensional data, and generating a weft texture and a warp texture based on the vector data; acquiring a current particle position, calculating to obtain a velocity vector based on the current particle position, the weft-wise texture and the warp-wise texture, and rendering the velocity vector into a velocity texture; calculating and rendering based on the current particle position and the speed texture to obtain a next particle position texture; generating and rendering based on the position texture and the speed texture of the next particle to obtain a post-processing position texture; and obtaining the position of a previous particle, calculating screen coordinates of the position of the previous particle, the position of the current particle, the post-processing position texture and the color band texture, and performing rendering processing on the screen coordinates to obtain a frame cache. According to the embodiment of the invention, rapid rendering of a large amount of NC data at a webpage end can be realized.

This invention discloses a lightweight remote sensing scene classification method based on frequency domain and multi-scale spatial features, belonging to the field of remote sensing image processing technology. This method aims to address the problems of existing deep learning models struggling to balance local details and global context when extracting features from remote sensing images, as well as their excessive computational complexity. The method mainly includes: employing a multi-branch parallel architecture to differentiate sub-feature groups; utilizing a channel grouping strategy to construct a four-way parallel architecture, performing original feature identity preservation, local spatial texture extraction, mid-range context-level fusion, and global frequency domain information enhancement respectively, thereby achieving complementarity and synergy of multi-scale spatial-frequency features; finally, the output features of each branch are concatenated and fused to obtain the final classification result. This invention, through spatial-frequency domain synergy and channel grouping strategies, significantly reduces the number of model parameters and computational load while effectively improving the feature capture capability and classification accuracy for complex remote sensing scenes.

This application relates to a method, system, and medium for extracting radio frequency (RF) signal features from unmanned aerial vehicles (UAVs). The method includes: decoupling the acquired RF signal of a target UAV using a time-frequency transformation algorithm to obtain a time-frequency distribution feature map; extracting spectral texture features from the time-frequency distribution feature map at different receptive field scales using a trained deep neural network model to obtain texture feature vectors; generating position feature vectors containing temporal logic information using the same deep neural network model; and performing a feature space fusion mapping between the texture feature vectors and the position feature vectors to generate the RF fingerprint features of the target UAV. This application achieves texture extraction based on dilated convolution and position perception based on self-attention. This dual-stream parallel network architecture simultaneously captures multi-scale spectral texture and long-range frequency hopping logic in the time-frequency domain, constructing a highly robust semantic fingerprint of RF signals and effectively improving the detection and recognition rate of UAV frequency hopping flight control signals.

The application discloses a kind of based on hyperspectral banana wilt leaf phenotype texture extraction method, especially in the early stage of disease leaf yellowing phenotype texture extraction method.The implementation function of the present application includes two parts of feature spectral image preprocessing, texture calculation.In feature spectral image preprocessing part, mainly for the feature spectral image of banana wilt related image preprocessing is carried out, the position of banana leaf is extracted from the image;Texture calculation part mainly carries out filtering, gradient calculation, different waveband image fusion etc.to the feature spectral image after extracting leaf position, finally obtains the real leaf phenotype texture information.Compared with the method of extracting texture features of ordinary color image, the texture features extracted by hyperspectral camera under two characteristic wavebands in the present application are fused, which can reflect the real banana leaf texture better than ordinary camera, and provide phenotype information support for subsequent visual recognition and disease judgment.

The invention discloses a double-mask multi-mode tongue picture analysis method and system fusing pulse condition prior. The method comprises the following steps: firstly, constructing an interactive double-flow trunk based on a convolutional neural network and Transform, and extracting multi-scale visual features; secondly, designing a double-mask feature decoupling module, filtering a background by using an inner mask to purify internal textures, and constructing an edge enhancement branch by using a large-scale expanded outer mask in cooperation with a Scharr operator to accurately capture high-frequency features such as tooth marks and cracks; then, constructing a pulse condition guiding module, introducing a numerical pulse condition vector through a FiLM mechanism, and performing channel-level dynamic modulation on the visual features; and finally, carrying out model optimization by adopting two-stage deep supervision course learning and a cosine annealing hot restart strategy. According to the method, the conflict between the edge features and internal texture extraction is effectively solved, the limitation of a single visual mode is overcome, and the recognition precision of the subtle morphological features of the tongue image is improved.

The application discloses a sea surface stereo matching and depth measurement method based on wave main direction correction, and aims to solve the feature degradation and mismatch caused by the consistency of the transverse sea wave texture and the matching search direction. The method first denoises and enhances the image and extracts the texture, calculates the global flow direction by using the block Hough transform, and adaptively starts the rotation optimization; secondly, an ambiguity scoring model based on normalized cross correlation (NCC) is constructed to filter the optimal rotation angle, the image is synchronously rotated and cropped, and the transverse repeated texture is reconstructed into a periodic structure with high gradient change; then, feature points are extracted in the rotated image domain, the RANSAC algorithm is combined to filter and estimate the fundamental matrix to complete the epipolar rectification, the prior disparity search range is locked according to the distribution of the homonymic points, and the semi-global matching (SGM) algorithm is used to generate the initial disparity; finally, the real depth is recovered through reverse correction, the point cloud is generated, and the three-dimensional surface of the sea wave is reconstructed. The application fundamentally eliminates the matching confusion caused by the transverse texture, significantly improves the three-dimensional reconstruction accuracy, and is suitable for application scenes such as marine wave monitoring and disaster warning.

The application discloses a flocculation state monitoring method based on texture features, which comprises the following steps: taking flocculation water as a sample, dividing the flocculation water into several equal parts, placing the flocculation water in transparent containers respectively, making the flocculation water in a suspended state, taking multiple sample images by using a camera, selecting sample images for pretreatment, extracting a gray level co-occurrence matrix of the images, calculating texture feature values of the images by using a texture extraction algorithm, taking average values of the texture feature values of multiple images of the same sample as results, repeating the above operations to obtain texture feature values of flocculation images in different time periods, calculating relative change amounts of the texture feature values respectively by using a formula, and using the relative change amounts of the texture feature values to judge the state of the flocculation.

The invention provides a three-dimensional map construction method and system for a transformer substation construction site, and relates to the technical field of three-dimensional maps, and the method comprises the steps: collecting point cloud data of the construction site through a laser radar, obtaining corresponding image data through an image collection device, carrying out the filtering, down-sampling and coordinate registration processing of the point cloud data, and obtaining a three-dimensional map; distortion correction and texture extraction are carried out on the image data; performing coordinate alignment on the point cloud data and the image data, identifying fusion features, and generating a three-dimensional model; integrating the structural data based on the three-dimensional model, and constructing structured three-dimensional update data; and adding the structured three-dimensional update data into the three-dimensional model, performing parameter verification on the three-dimensional map data, and fusing the verified three-dimensional map parameters with the three-dimensional model to generate a three-dimensional map. According to the method and the device, the technical problem of relatively poor construction accuracy of the three-dimensional map in the prior art can be solved, and the technical effect of improving the construction accuracy of the three-dimensional map is achieved.

The present application provides a kind of based on near infrared remote sensing texture feature's sugar beet yield prediction method and system, it is related to sugar beet yield prediction technical field, the present application obtains monitoring area remote sensing image and extracts near infrared band data, generates standard gray texture map by nonlinear stretching, effectively enhances crown layer texture details, subsequently traverses full map using spatial coexistence constraint condition, constructs omnidirectional gray coexistence probability matrix eliminating directionality difference, solves the problem of being limited by scanning direction and image size when extracting traditional texture, and then extracts leaf broken degree representing the degree of leaf edge overlap and local cohesion representing the tightness of crown layer growth, from spatial distribution dimension, it is sensitive to capture the micro feature of crop high biomass area, finally, generate crown layer shrinkage index using exponential decay mechanism, and substitute into nonlinear inversion equation to output sugar beet predicted yield.