Multi-band remote sensing image crop identification method, device, equipment and medium

Abstract

The invention relates to the field of image recognition, can be applied to business system platforms of financial science and technology, medical treatment and health and the like, and discloses a multiband remote sensing image crop recognition method, device and equipment and a medium. Performing standardization processing on the multi-temporal time sequence data sequence, extracting a pixel neighborhood of standard image data, generating a plurality of pixel data blocks according to the pixel neighborhood, splicing the pixel data blocks to obtain a time sequence feature image, performing random shielding on the obtained standard strip-shaped image, and performing data enhancement on a preset feature extraction model to obtain a time sequence feature image; and obtaining a crop identification model, carrying out convolution on the time sequence feature image by using the obtained crop type information and the crop identification model to obtain an image crop type, and generating a plot crop distribution map according to the image crop type. According to the method, the precision, adaptability and stability of multi-band remote sensing image crop identification are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image recognition, and in particular to a multi-band remote sensing image crop identification method, device, equipment and medium. BACKGROUND

[0002] In the prior art, in the field of agricultural insurance, the existing technology and products for crop identification have been applied to a certain extent, but most of them still have many shortcomings and are difficult to meet the needs of modern agricultural fine management and precise insurance claims. The common crop identification technology in the industry currently mainly relies on remote sensing image processing and simple classification algorithms, but these methods have significant limitations in dealing with complex natural environments and diverse crop species.

[0003] In the field of medical health, although the current medical image recognition and disease prediction system has been widely applied, there are still technical shortcomings similar to crop identification. For example, most systems still rely on a single modality (such as X-ray or MRI) for diagnosis, lack of multi-modal fusion (such as combining physiological parameters, medical records, laboratory tests, etc.), resulting in low recognition accuracy for complex conditions. Many methods are still based on static images, making it difficult to dynamically track the development process of lesions, limiting the ability to continuously assess the evolution of chronic diseases or tumors.

[0004] In the field of financial technology business, for example, in the field of financial risk control and intelligent credit, existing models rely on structured historical data (such as transaction records, credit scores, etc.) and simple rule models for risk identification, lack of fusion analysis of multi-dimensional behavior characteristics (such as real-time consumption behavior, social signals, geographic location, etc.), resulting in slow response to fraud methods and user credit changes. At the same time, the model is often based on a static snapshot and cannot perform dynamic evolution analysis, such as the time series changes in user credit trends or risk exposure, affecting the forward-looking nature of the decision.

[0005] In summary, the existing technology relies too much on single-band and NDVI images, does not fully utilize multi-band information, has low recognition accuracy; uses static image processing, lacks dynamic monitoring of crop growth cycles; has poor adaptability in complex environments and climate changes; the model is generally shallow, with weak feature extraction and generalization ability; lacks effective compensation and correction mechanisms for data missing and noise interference, affecting the stability and reliability of the identification results.

[0006] Therefore, in the current technology, there are problems of low precision, poor adaptability, and low stability in multi-band remote sensing image crop identification. SUMMARY

[0007] The present application provides a multi-band remote sensing image crop identification method, device, equipment and medium, which mainly aims to solve the problems of low precision, poor adaptability and low stability in multi-band remote sensing image crop identification.

[0008] In a first aspect, to achieve the above object, the present application provides a multi-band remote sensing image crop identification method, comprising: acquiring multi-band remote sensing image data, generating a multi-temporal time series data sequence according to the multi-band remote sensing image data, and performing standardization processing on the multi-temporal time series data sequence to obtain standard image data; extracting a pixel neighborhood of the standard image data, and generating a plurality of pixel data blocks according to the pixel neighborhood; splicing the pixel data blocks to obtain a time series feature image; acquiring a standard bar image, randomly masking the standard bar image, and using the masked standard bar image to perform data enhancement on a preset feature extraction model to obtain a crop identification model; acquiring crop category information, convolving the time series feature image using the crop category information and the crop identification model to obtain an image crop category; generating a field crop distribution map according to the image crop category.

[0009] In a second aspect, the present application further provides a multi-band remote sensing image crop identification device, comprising: an image data standardization module configured to acquire multi-band remote sensing image data, generate a multi-temporal time series data sequence according to the multi-band remote sensing image data, and perform standardization processing on the multi-temporal time series data sequence to obtain standard image data; a pixel data block generation module configured to extract a pixel neighborhood of the standard image data, and generate a plurality of pixel data blocks according to the pixel neighborhood; a pixel data block splicing module configured to splice the pixel data blocks to obtain a time series feature image; an identification model optimization module configured to acquire a standard bar image, randomly mask the standard bar image, and use the masked standard bar image to perform data enhancement on a preset feature extraction model to obtain a crop identification model; a feature image convolution module configured to acquire crop category information, convolve the time series feature image using the crop category information and the crop identification model to obtain an image crop category; a crop distribution map generation module configured to generate a field crop distribution map according to the image crop category.

[0010] In a third aspect, the present application further provides an electronic device, comprising: at least one processor; and a memory in communication with the at least one processor; wherein The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the multi-band remote sensing image crop identification method described above.

[0011] In a fourth aspect, the present application further provides a computer readable storage medium, wherein at least one computer program is stored in the computer readable storage medium, and the at least one computer program is executed by a processor in an electronic device to implement the multi-band remote sensing image crop identification method described above.

[0012] The present application obtains multi-band remote sensing image data, generates a multi-temporal time series data sequence according to the multi-band remote sensing image data, and performs standardization processing on the multi-temporal time series data sequence to obtain standard image data, which not only significantly improves the quality and comparability of the data, but also provides a stable and reliable data basis for subsequent crop identification, change detection, time series analysis and other remote sensing applications. The pixel neighborhood of the standard image data is extracted, and a plurality of pixel data blocks are generated according to the pixel neighborhood, effectively fusing local spatial structure and multispectral information, enhancing the data expression capability, retaining the details of the pixels and the neighborhood correlation, and helping to improve the accuracy of subsequent classification and identification. The pixel data blocks are spliced to obtain a time series feature image. By converting the multi-band fused pixel data blocks into a two-dimensional pixel image and splicing along the horizontal direction, combining linear stretching and normalization processing, the spatial and spectral features of multi-temporal and multi-sample can be effectively integrated and standardized, the contrast and detail performance of the time series feature image are enhanced, a standard bar image is obtained, the standard bar image is randomly masked, and the masked standard bar image is used to perform data augmentation on a pre-set feature extraction model to obtain a crop identification model. The crop identification model effectively simulates the occlusion, noise and environmental interference in the remote sensing image, improves the diversity and complexity of the training data, obtains crop type information, and convolves the time series feature image using the crop type information and the crop identification model to obtain an image crop type. By extracting the spectral difference, dynamic change and spatial form features of crops in the three dimensions of band, time and space respectively, and fusing them into high-dimensional convolution features for pooling compression, further combining the class encoding of the crop type, and obtaining the probability distribution of each crop type through full connection operation, the class corresponding to the maximum probability is finally taken as the recognition result. Not only does it make full use of the multi-dimensional information of remote sensing data, but also introduces class prior knowledge to improve the model discrimination ability, realizes accurate and efficient identification of crop types, significantly improves the generalization performance and classification accuracy of the model in complex environments, generates a field crop distribution map according to the image crop type, ensures the completeness of the remote sensing recognition result in space and the rationality of the real field boundary, and improves the precision, adaptability and stability of multi-band remote sensing image crop identification. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments of the present application will be briefly introduced as follows. Obviously, the drawings described in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0014] Figure 1 An application environment schematic diagram of a multi-band remote sensing crop identification method according to an embodiment of the present application; Figure 2 A flow schematic diagram of a multi-band remote sensing crop identification method according to an embodiment of the present application; Figure 3 A flow schematic diagram of a crop distribution map generation module in a multi-band remote sensing crop identification method according to an embodiment of the present application; Figure 4 A module schematic diagram of a multi-band remote sensing crop identification device according to an embodiment of the present application; Figure 5 A structure schematic diagram of an electronic device for implementing a multi-band remote sensing crop identification method according to an embodiment of the present application; Figure 6 Another structure schematic diagram of an electronic device for implementing a multi-band remote sensing crop identification method according to an embodiment of the present application.

[0015] The object implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0016] In order to make the person skilled in the art better understand the technical solutions of the present disclosure, and to fully understand and implement the implementation process of the present disclosure how to apply technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of the present disclosure will be described clearly and completely in the embodiments of the present disclosure with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present disclosure, not all embodiments. The embodiments of the present disclosure and each feature in the embodiments can be combined with each other without conflict, and the technical solutions formed thereby are all within the protection scope of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present disclosure.

[0017] It should be noted that the terms "first", "second", and the like in the description and claims of the present disclosure and above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, device, product or apparatus including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or apparatuses.

[0018] The embodiment of the present application provides a kind of multi-band remote sensing map crop identification method, the execution subject of the multi-band remote sensing map crop identification method includes but is not limited to at least one of the electronic equipment that can be configured to execute the device provided by the embodiment of the present application, such as server, terminal etc.It is said that the multi-band remote sensing map crop identification method can be executed by the software or hardware installed in terminal equipment or server equipment.The server includes but is not limited to: single server, server cluster, cloud server or cloud server cluster etc.The server can be independent server, can also be cloud server that provides cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, content distribution network (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platform.

[0019] The embodiment of the present application provides a kind of multi-band remote sensing map crop identification method, which can be applied to Figure 1In the application environment, the client communicates with the server through the network. The server can obtain multi-band remote sensing image data through the client, generate a multi-temporal time series data sequence according to the multi-band remote sensing image data, and perform standardization processing on the multi-temporal time series data sequence to obtain standard image data. Not only does it significantly improve the quality and comparability of the data, but also provides a stable and reliable data foundation for subsequent crop identification, change detection, time series analysis and other remote sensing applications. The pixel neighborhood of the standard image data is extracted, and a plurality of pixel data blocks are generated according to the pixel neighborhood. The local spatial structure and multispectral information are effectively fused, and the data expression capability is enhanced. Not only the details of the pixels and the neighborhood correlation are retained, but also the accuracy of subsequent classification and identification is improved. The pixel data blocks are spliced to obtain a time series feature image. By converting the multi-band fused pixel data blocks into a two-dimensional pixel image and splicing along the horizontal direction, combining linear stretching and normalization processing, the spatial and spectral features of multi-temporal and multi-sample can be effectively integrated and standardized, and the contrast and detail performance of the time series feature image are enhanced. The standard bar image is obtained, the standard bar image is randomly masked, and the masked standard bar image is used to perform data augmentation on the preset feature extraction model to obtain a crop identification model. The occlusion, noise and environmental interference in the remote sensing image are effectively simulated, and the diversity and complexity of the training data are improved. The crop type information is obtained, and the crop type information and the crop identification model are used to convolve the time series feature image to obtain an image crop type. By extracting the spectral difference, dynamic change and spatial form features of the crop in the three dimensions of band, time and space respectively, and fusing them into high-dimensional convolution features for pooling compression, the class encoding of the crop type is further combined. Through full connection operation, the probability distribution of each crop type is obtained, and finally the class corresponding to the maximum probability is taken as the recognition result. Not only does it make full use of the multi-dimensional information of remote sensing data, but also introduces class priori knowledge to improve the model discrimination ability, realizes accurate and efficient identification of crop types, significantly improves the generalization performance and classification accuracy of the model in complex environments, generates a plot crop distribution map according to the image crop type, ensures the integrity of the remote sensing recognition result in space and the rationality of the real plot boundary, improves the precision, adaptability and stability of the multi-band remote sensing crop identification, and finally outputs the plot crop distribution map back to the user client. The client can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices. The server can be implemented by an independent server or a server cluster composed of multiple servers. The application will be described in detail below through specific embodiments.

[0020] The following is explained in the description of the present application. The present application extracts the spectral difference, dynamic change and spatial morphological characteristics of crops in the three dimensions of waveband, time and space respectively, fuses them into high-dimensional convolution features after pooling and compression, further combines the class encoding of crop categories, obtains the probability distribution of each crop category through full connection operation, and finally takes the category corresponding to the maximum probability as the recognition result. Not only does it make full use of the multi-dimensional information of remote sensing data, but also introduces the prior knowledge of categories to improve the discrimination ability of the model, realizes the accurate and efficient recognition of crop categories, and significantly improves the generalization performance and classification accuracy of the model in complex environments.

[0021] Referring to Figure 2 Fig. 1 shows a flowchart of a multi-band remote sensing image crop recognition method according to an embodiment of the present application. In this embodiment, the multi-band remote sensing image crop recognition method comprises: S1, acquiring multi-band remote sensing image data, generating multi-temporal time series data sequences according to the multi-band remote sensing image data, and performing standardization processing on the multi-temporal time series data sequences to obtain standard image data.

[0022] In the embodiment of the present application, multi-band remote sensing image data is acquired and sorted according to its time sequence to construct a multi-temporal time series data sequence. The sequence is sequentially subjected to radiation calibration, atmospheric correction, geometric correction and cloud mask processing, and finally a labeled time sequence is obtained. The labeled sequence is subjected to resolution uniform processing to ensure that each temporal image has consistent spatial resolution, thereby obtaining standardized multi-temporal remote sensing image data as the basis for subsequent analysis.

[0023] In the specific scenario of medical health, a similar multi-temporal image data processing flow can be applied to public health monitoring and environmental health analysis. For example, by acquiring multi-temporal atmospheric pollution remote sensing data in a certain area, combining time series for radiation calibration, atmospheric and geometric correction, marking high pollution periods or areas, and uniformly processing the resolution, standardized pollution time series image data can be generated for analyzing the correlation between pollution diffusion trend and residents' health events (such as respiratory diseases), thereby assisting the government in disease warning and health intervention strategy formulation.

[0024] In the specific scenario of financial technology, it can be used for agricultural risk control and credit assessment. For example, remote sensing image data of agricultural areas is acquired, and after multi-temporal standardization processing, the crop categories and growth states can be accurately identified, and the existence of disaster or yield reduction risk can be judged according to the time series change trend. Financial institutions can dynamically assess the credit risk of agricultural loan objects or provide objective basis for insurance claims, thereby improving the automation and accuracy of risk control.

[0025] In the embodiment of the present application, the multi-band remote sensing image data is acquired, a multi-temporal time series data sequence is generated according to the multi-band remote sensing image data, and the multi-temporal time series data sequence is standardized to obtain standard image data, comprising: The time sequence of the multi-band remote sensing image data is acquired, and the multi-band remote sensing image data is sorted according to the time sequence to obtain a multi-temporal time series data sequence; The multi-temporal time series data sequence is radiometrically calibrated to obtain a calibrated time series; The calibrated time series is subjected to atmospheric correction and geometric correction to obtain a corrected time series; The cloud shadow coverage area of the corrected time series is marked to obtain a marked time series; The marked time series is subjected to resolution uniformity processing to obtain standard image data.

[0026] In detail, multi-band remote sensing image data covering a target area is acquired, and the acquisition time information of each image is extracted. The time sequence of all images is sorted according to the acquisition time, and a multi-temporal time series data sequence reflecting the change of the ground target over time is constructed, providing basic data support for subsequent time series analysis and change detection.

[0027] The digital number (DN value) in the original remote sensing image is converted into a radiation brightness or apparent reflectance with physical meaning, so as to eliminate the inconsistency caused by the difference in sensor performance or the change in observation conditions. Usually, according to the calibration coefficient or metadata provided by the sensor, each time phase image is processed by linear transformation and other methods, so as to obtain a calibrated time series with comparability, laying a foundation for subsequent atmospheric correction and quantitative analysis.

[0028] Atmospheric correction and geometric correction of the calibrated time series are performed to further improve the authenticity and spatial consistency of the remote sensing image. Atmospheric correction converts the apparent reflectance of the image into the real reflectance of the ground object by removing the influence of atmospheric scattering and absorption, ensuring the spectral comparability between different time phase images. Geometric correction calibrates the spatial position deviation of the image, unifies each time phase image to the same geographic coordinate system, and eliminates the spatial misplacement caused by the sensor viewing angle, orbit drift or terrain undulation, thereby generating a corrected time series with spatial alignment and spectral consistency.

[0029] Using multi-band information and spatio-temporal characteristics, the area in the remote sensing image blocked by the cloud layer or its shadow is identified and marked in the image. Usually, combined with a cloud detection algorithm (such as based on spectral threshold, temperature difference or machine learning method), the spatial distribution of the cloud layer and the shadow is extracted, and the corresponding mask layer is generated to mark the pixel position that cannot be used or needs to be excluded, thereby obtaining a marked time series, which provides a basis for subsequent data cleaning, interpolation completion or analysis modeling.

[0030] Adjusting remote sensing images at different time nodes to a consistent spatial resolution to ensure the comparability of each phase data in the spatial scale usually includes upsampling (such as bilinear interpolation, cubic convolution interpolation) of low-resolution images or downsampling of high-resolution images to match the preset unified resolution standard. In the processing process, the spatial structure characteristics of the image need to be preserved and the interpolation error needs to be minimized, and finally the standard image data with unified spatial accuracy and complete time label can be directly used for subsequent time series analysis, classification identification or feature change detection and other tasks.

[0031] Converting the original multi-band remote sensing image data into standardized multi-temporal image data with unified time sequence, radiation consistency, spatial alignment, cloud shadow information marking and unified resolution not only significantly improves the quality and comparability of the data, but also provides a stable and reliable data foundation for subsequent crop identification, change detection, time series analysis and other remote sensing applications, effectively improves the accuracy and robustness of model analysis, reduces the interference of data preprocessing on the analysis results, and enhances the practical value of remote sensing technology in the fields of agriculture, ecology, environment and the like.

[0032] S2, extracting a pixel neighborhood of the standard image data, and generating a plurality of pixel data blocks according to the pixel neighborhood.

[0033] In the embodiment of the application, a plurality of sample center pixel points are randomly selected from the standard image data, a pixel neighborhood containing spatial context is formed according to each sample center pixel point, the numerical values of all pixels in each neighborhood at each band are obtained, and a three-dimensional pixel data block fusing spatial and spectral features is constructed, thereby effectively preserving the local spatial structure and multi-spectral information of the center pixel and providing rich and structured input data for subsequent classification identification, feature extraction or deep learning model training.

[0034] In the medical health specific scene, it can be applied to medical image analysis. For example, when processing high-resolution MRI or CT images, center pixel points can be selected from key lesion areas, image blocks containing neighborhoods are constructed, intensity information of each pixel in different imaging sequences (such as T1, T2, FLAIR, etc.) is extracted, and three-dimensional image block data fusing spatial structure and multi-modal features are formed. These structured samples can be used to train deep learning models to achieve high-precision identification and classification of tumors, lesions or tissue abnormalities, and improve the accuracy and intelligent level of medical auxiliary diagnosis.

[0035] In the specific scenario of financial technology, remote sensing assisted credit risk assessment and asset monitoring can be used. For example, in the agricultural loan scenario, representative pixels of farmland areas are selected from standardized remote sensing images, image blocks containing spatial neighborhoods are constructed, and spatial layout and multi-temporal spectral features of crops are extracted as criteria for crop growth, planting density and health status. Such structured sample data can be used to train risk control models to assist financial institutions in determining planting conditions and expected production capacity, improving the precision management and risk control capability of agricultural credit.

[0036] In the embodiment of the present application, the pixel neighborhood of the standard image data is extracted, and a plurality of pixel data blocks are generated according to the pixel neighborhood, comprising: Randomly selecting a plurality of sample center pixel points from the standard image data, and generating a two-dimensional neighborhood window of a predetermined size according to the sample center pixel points one by one; Extracting a pixel neighborhood centered on the sample center pixel point on the standard image data according to the size of the two-dimensional neighborhood window; Obtaining all band values in each pixel neighborhood, stacking the band values in a predetermined vertical direction to obtain a pixel data block.

[0037] In detail, a plurality of sample center pixel points are randomly selected from the standard image data, a two-dimensional neighborhood window of a predetermined size is constructed around each center pixel point, all pixel information within the window is extracted to form a local area containing the center pixel and its surrounding context, and rich spatial structure data support is provided for subsequent feature extraction and analysis.

[0038] On the standard image data, a pixel set within a corresponding range around each sample center pixel point is extracted according to a predetermined two-dimensional neighborhood window size to form a pixel neighborhood containing spatial context information, ensuring that the pixels in the neighborhood can reflect the local features of the area where the center pixel is located, and laying a foundation for subsequent multi-band feature fusion and analysis.

[0039] For each pixel in the two-dimensional neighborhood window, the corresponding pixel value is extracted wave by wave, and the two-dimensional pixel matrix of each band is stacked into a three-dimensional data block according to the band order, wherein the first dimension and the second dimension correspond to the row and column coordinates of the space, and the third dimension corresponds to the spectral dimension of different bands, i.e. for each position pixel in the neighborhood, the values are arranged into a spectral vector in the order of bands, and all position spectral vectors are combined to form a complete three-dimensional pixel data block, thereby realizing the fusion of spatial information and spectral information.

[0040] By randomly selecting sample center pixels from standard image data and constructing a two-dimensional neighborhood window, a pixel neighborhood containing spatial context is extracted, and then the pixel values of each band are stacked according to the band dimension to form a three-dimensional pixel data block, which effectively fuses local spatial structure and multispectral information, enhances the expression ability of the data, not only retains the detail features and neighborhood correlation of the pixels, which helps to improve the accuracy of subsequent classification and recognition, but also provides rich and structured input data for complex models such as deep learning, and improves the discrimination ability and robustness of the model to target objects.

[0041] S3, splicing the pixel data block to obtain a time sequence feature image.

[0042] In the embodiment of the application, each pixel data block is converted into a two-dimensional pixel image, and then the two-dimensional images are spliced in sequence according to the preset horizontal direction to form a continuous spliced pixel image. The spliced image is linearly stretched to enhance the contrast and detail performance of the image. Finally, the pixel value range is adjusted through normalization processing to obtain a time sequence feature image with uniform scale and good visual features, providing high-quality input for subsequent time sequence analysis and model training.

[0043] In the medical health specific scenario, it can be used for processing and analysis of multi-phase medical image data. By fusing and splicing the spatial and spectral features of multi-modal images (such as MRI, CT, PET, etc.) obtained at different time points, a unified time sequence feature image is formed, which helps doctors to intuitively observe the trend of lesion changes over time, supports disease progression monitoring, efficacy evaluation and personalized treatment plan formulation, and improves the accuracy and timeliness of diagnosis.

[0044] In the financial technology specific scenario, it can be used for agricultural financial risk assessment and asset monitoring. By splicing multi-band remote sensing image data of different periods into a standardized time sequence feature image, the growth state and trend of crops are accurately reflected, helping financial institutions to realize crop growth monitoring, disaster warning and yield prediction based on remote sensing data, improve the scientific nature and efficiency of loan approval and insurance claims, and reduce financial risks.

[0045] In the embodiment of the application, the splicing of the pixel data block to obtain a time sequence feature image comprises: converting the pixel data block into a two-dimensional pixel image; splicing the two-dimensional pixel image in a preset horizontal direction to obtain a spliced pixel image; linearly stretching the spliced pixel image to obtain a stretched pixel image; normalizing the stretched pixel image to obtain a time sequence feature image.

[0046] In detail, the three-dimensional pixel data block is unfolded or rearranged in the wave band dimension, the spectral information of multiple wave bands is mapped to different channels or positions of the two-dimensional image, so that the three-dimensional data originally containing spatial and spectral characteristics is expressed in the form of a two-dimensional image, facilitating subsequent analysis and feature extraction using two-dimensional image processing technology, while maintaining the spatial structure of the pixels and the integrity of the spectral information.

[0047] The two-dimensional pixel images are sequentially arranged and connected in a preset horizontal direction, and multiple pixel images are seamlessly spliced along the horizontal direction to form a wide spliced pixel image, maintaining the spatial structure continuity of each image and realizing the horizontal expansion of multi-sample or multi-temporal pixel features, thereby providing structured input data for subsequent unified processing and analysis.

[0048] The image pixel values are mapped from the original range to a preset target range (such as 0 to 255 or 0 to 1) through linear transformation to enhance the contrast and detail performance of the image, the stretching ratio and offset are calculated according to the minimum and maximum values of the pixels in the image, and the linear mapping is applied to each pixel to make the pixel value distribution more uniform, improve the visual effect of the image and the sensitivity of the subsequent algorithm to features, and obtain a stretched pixel image with better recognition.

[0049] The pixel values are linearly converted according to a predetermined range (usually 0 to 1 or -1 to 1) to eliminate the scale difference of pixel values between different images, ensure that the data has a uniform numerical range and distribution characteristics, and help improve the stability and convergence speed of subsequent model training, while avoiding calculation bias caused by too large numerical difference, and finally obtain standardized time series feature images suitable for time series analysis and deep learning model input.

[0050] By converting the multi-waveband fused pixel data block into a two-dimensional pixel image and splicing along the horizontal direction, combined with linear stretching and normalization processing, the spatial and spectral features of multi-temporal and multi-sample can be effectively integrated and standardized, and the data structure and expression ability can be improved. This processing method not only enhances the contrast and detail performance of the time series feature image, but also ensures the numerical consistency of the data, which is beneficial to the training and inference of subsequent deep learning models, improves the accuracy and stability of classification and recognition, and promotes the efficient use and intelligent analysis of complex time series remote sensing data.

[0051] S4, obtaining a standard strip image, randomly masking the standard strip image, and using the masked standard strip image to perform data augmentation on a preset feature extraction model to obtain a crop recognition model.

[0052] In the embodiment of the present application, the standard bar image is randomly masked according to the set masking proportion interval, the image is divided into several initial masking area blocks, one of the area blocks is randomly selected, Gaussian random noise is added to the area block, the sample area image containing noise is used to perform data enhancement and optimization training on the preset feature extraction model, the robustness and generalization ability of the model are improved, and a more accurate and stable crop recognition model is obtained.

[0053] In the medical health specific scenario, it can be used for medical image data enhancement. By randomly masking and injecting Gaussian noise into the lesion area, the image blur, occlusion or noise interference that may exist in the real clinical environment is simulated, thereby expanding the diversity of training samples and improving the recognition robustness of the model to abnormal tissues or lesions, thereby assisting early diagnosis and precise treatment of diseases.

[0054] In the financial technology specific scenario, it can be used for remote sensing image assisted agricultural risk assessment. By randomly masking and injecting noise into the key area of the crop growth image, the sensor error or cloud cover and other uncertain factors are simulated, the adaptability of the model to crop state recognition under diversified environmental conditions is enhanced, and the stability and accuracy of the agricultural insurance claim and credit risk control model are improved.

[0055] In the embodiment of the present application, the standard bar image is randomly masked, and the masked standard bar image is used for data enhancement of the preset feature extraction model to obtain a crop recognition model, comprising: An interval of masking proportions is obtained, and the standard bar image is masked according to the interval of masking proportions to obtain a plurality of initial masking area blocks; One of the initial masking area blocks is randomly selected as a target area block; Gaussian random noise is added to the target area block to obtain a sample area image; The sample area image is used to optimize the preset feature extraction model to obtain a crop recognition model.

[0056] In detail, according to the preset masking proportion interval, part of the area of the standard bar image is masked in a random or regular manner, the image is divided into a plurality of initial masking area blocks alternately appearing in the masked and unmasked states, thereby forming a segmented image containing a variety of masking modes, and providing diversified samples for subsequent data enhancement and model training.

[0057] One of the initial masking area blocks is randomly selected as a target area block, and Gaussian random noise is applied to the target area block, that is, random noise conforming to Gaussian distribution is superimposed in the pixel value to simulate the sensor noise or environmental interference that may occur in reality, a sample area image with noise is generated, and is used to enhance the robustness of the model to noise and uncertainty.

[0058] The noise-containing sample region image is used as enhanced data to input a preset feature extraction model, the learning experience of the model is enriched by diversified training samples, and the adaptability of the model to noise, shielding and environmental changes in the image is improved. After repeated iterative training, the model gradually enhances the recognition and differentiation ability of crop features in complex scenes, and finally a crop recognition model with strong robustness and high recognition accuracy is obtained, which significantly improves the application effect in actual remote sensing images.

[0059] By randomly shielding the standard bar image and adding Gaussian random noise in the selected region block, the shielding, noise and environmental interference in the remote sensing image are effectively simulated, and the diversity and complexity of the training data are improved. Using these enhanced samples to optimize the feature extraction model not only enhances the robustness of the model to abnormal and incomplete data, but also improves the accuracy and stability of crop recognition, significantly improving the adaptability and generalization performance of the model in complex practical application environments.

[0060] S5, obtaining crop category information, and performing convolution on the time sequence feature image using the crop category information and the crop recognition model to obtain image crop categories.

[0061] In the embodiment of the application, the time sequence feature image is subjected to multi-layer convolution operation along the three dimensions of waveband, time and space in sequence by using the crop recognition model, rich information such as waveband spectral difference, dynamic change feature and crop growth form is extracted, and these multi-dimensional convolution features are summarized to form high-dimensional feature representation. The high-dimensional features are reduced in dimension by using average pooling, combined with the class coding of crop category information, the spatial, spectral and temporal features and the class prior are fused through the full connection layer, and finally the probability distribution of each crop class is calculated. The class corresponding to the maximum probability is taken as the crop recognition result of the time sequence feature image, and the precise classification of crop categories is realized.

[0062] In the specific scenario of medical health, the multi-dimensional convolution feature extraction and fusion technology can be applied to multi-modal medical image analysis. By performing convolution processing along the spectral, temporal and spatial dimensions on different imaging sequences (such as T1, T2 and diffusion weighted imaging of MRI), tissue structure, lesion dynamic change and morphological features are extracted, combined with the pathological class prior, the precise recognition and classification of tumor types and lesion progression are realized, which assists doctors in formulating individualized diagnosis and treatment plans, and improves the accuracy and efficiency of diagnosis.

[0063] In the specific scenario of financial technology, remote sensing image assisted agricultural risk assessment and credit management can be used. By multi-dimensional convolution, the multi-temporal spectral features and crop growth morphology of farmland are extracted, the crop type prior information is fused, and the accurate discrimination of crop type and growth condition is realized. Based on this, financial institutions can dynamically monitor agricultural production risks, optimize loan approval processes and insurance claim decision-making, and improve risk control capabilities and fund use efficiency.

[0064] In the embodiment of the application, the convolution of the time sequence feature image by the crop type information and the crop recognition model obtains an image crop type, which includes: The wave band dimension of the time sequence feature image is subjected to first layer convolution by the crop recognition model to obtain wave band spectral difference; The time dimension of the time sequence feature image is subjected to second layer convolution to obtain dynamic change features; The spatial dimension of the time sequence feature image is subjected to third layer convolution to obtain crop growth morphology; The wave band spectral difference, the dynamic change features and the crop growth morphology are summarized as high-dimensional convolution features; The high-dimensional convolution features are subjected to average pooling to obtain high-dimensional pooling features; The crop type information is subjected to class encoding to obtain crop type encoding; The high-dimensional pooling features and the crop type encoding are subjected to full connection processing to obtain crop type probability; The crop type corresponding to the maximum crop type probability is taken as the image crop type.

[0065] In detail, under the premise of keeping the image spatial structure unchanged, a plurality of one-dimensional convolution kernels are used to slide along the wave band direction to model and extract the spectral response relationship between different wave bands, which can effectively capture the reflection characteristic difference of crops under different wave bands, extract spectral information representing physical and physiological characteristics of crops, and thus generate a feature map reflecting the spectral difference between wave bands, thereby providing a spectral basis for subsequent extraction of time sequence and spatial features.

[0066] After the spectral features are extracted, the convolution kernel is used to slide along the time axis direction to model and extract the spectral changes of the same spatial position at different time phases, which can effectively capture the dynamic change features of crops in the growth cycle, such as the rising and falling trend of reflectivity with time, the time sequence mode of growth stage, etc., and extract important time sequence features representing the growth process of crops, thereby providing key dynamic information support for final type recognition.

[0067] On the basis of extracting the spectrum and timing features, the spatial texture, structure distribution and morphological features of the crop region are modeled and extracted by sliding the two-dimensional convolution kernel in the spatial plane (i.e. row and column dimensions) of the image, which can capture the spatial distribution pattern, crown structure or arrangement form of the crop in the remote sensing image, thereby refining the high-order spatial features representing the growth form of the crop, providing supplementary information about the crop ground form for subsequent classification, and improving the accuracy and discrimination of recognition.

[0068] The band spectrum difference, dynamic change feature and crop growth form feature extracted through three-layer convolution operations are fused in the channel dimension to form a unified high-dimensional convolution feature representation, comprehensively integrating the spectral, temporal and spatial information, calculating the average value of the feature value in each local region, thereby realizing dimension reduction and compression of the features, retaining the global structure information while reducing redundancy, improving the compactness of feature expression and the computational efficiency of the model, and finally obtaining high-dimensional pooling features with representativeness and discrimination.

[0069] The original literal or symbolic crop name (such as "corn", "rice", etc.) is converted into a numerical form that the model can process. One-Hot Encoding or Embedding can be used to represent each crop as a unique vector, thereby obtaining the crop category code for model training or feature fusion, which not only retains the category distinction, but also provides a structured input for subsequent fusion with image features, helping to improve the model's understanding and utilization of category priors.

[0070] The image features fused with spectral, temporal and spatial information and the prior encoding of crop categories are jointly input into a fully connected neural network, which discriminates different crop categories through weight learning, outputs the probability distribution of each crop category, and selects the category with the maximum probability value as the corresponding crop category of the image, realizing automatic recognition and accurate classification of crop types in remote sensing images.

[0071] By extracting the spectral difference, dynamic change and spatial form features of the crop in the three dimensions of band, time and space, and fusing them into high-dimensional convolution features for pooling and compression, and further combining the category encoding of the crop, the probability distribution of each crop is obtained through fully connected operation, and finally the category corresponding to the maximum probability is taken as the recognition result. Not only does it make full use of the multi-dimensional information of remote sensing data, but also introduces category prior knowledge to improve the model's discrimination ability, realizes accurate and efficient recognition of crop categories, significantly improves the generalization performance and classification accuracy of the model in complex environments, and is suitable for large-scale and automated remote sensing crop recognition tasks.

[0072] S6、According to the image crop category, a plot crop distribution map is generated.

[0073] In the embodiment of the present application, according to the pixel position corresponding to the image crop category recognition result, the image crop category processed by block is spliced in space, and an initial crop distribution map covering the target area is reconstructed. The initial image edge area is extended and processed by Gaussian blur to reduce local noise and boundary mutation, and a more continuous crop distribution map is obtained. The adjacent same type area is further integrated by a region consistency enhancement algorithm to improve the coherence and integrity of the crop type distribution, thereby generating a clear, stable and actual plot boundary feature plot crop distribution map.

[0074] In the medical health specific scene, it can be used in multi-region lesion identification and organ structure reconstruction tasks. The lesion distribution map of the whole medical image is reconstructed by splicing the lesion area obtained by block identification in space. The edge area is extended and processed by Gaussian blur to reduce the artifacts and edge breakage in the identification result. Finally, the adjacent same type tissue area is integrated by region consistency enhancement to improve the coherence and structure clarity of the lesion identification result, thereby generating a fine lesion image with anatomical boundary features, providing more reliable auxiliary diagnosis basis for doctors.

[0075] In the financial technology specific scene, it can be used for agricultural asset atlas reconstruction based on remote sensing images. The crop type image identified by block is spliced into a complete area map to complete the boundary missing or cloud cover area in the remote sensing image, and the continuity and classification accuracy of the image are improved by using Gaussian blur and consistency enhancement algorithm processing, thereby generating a crop plot map with clear spatial structure and accurate regional distribution. This atlas can be used as an important basis for agricultural credit, insurance pricing and risk assessment, and can improve the precision and automation level of financial services.

[0076] Figure 3 A flowchart of a crop distribution map generation module in a multi-band remote sensing image crop identification method provided by an embodiment of the present application is shown.

[0077] In the embodiment of the present application, the generating a plot crop distribution map according to the image crop category includes: Obtaining the pixel position corresponding to the image crop category, splicing and restoring the image crop category according to the pixel position to obtain an initial crop distribution map; Edge extension is performed on the initial crop distribution map to obtain an extended crop distribution map; The extended crop distribution map is processed by Gaussian blur to obtain a smoothed crop distribution map; The region consistency of the smoothed crop distribution map is enhanced to obtain a plot crop distribution map.

[0078] In detail, according to the image crop type result identified by the model, the pixel position index information corresponding to each classification result is extracted, and these classification results are spliced and restored according to their corresponding positions according to the spatial arrangement rule of the original image, so that an initial crop distribution map covering the whole area is reconstructed, and the transition from the block recognition result to the regional crop type map is realized, thereby laying a foundation for subsequent boundary correction and distribution map optimization.

[0079] The edge region of the initial crop distribution map is extended, the pixel value of the edge region is extended to the outside of the image by copying or mirroring the edge pixel information, the boundary loss caused by splicing and cutting is filled, and a spatially continuous and edge complete extended crop distribution map is generated, so that the integrity and consistency of the edge information during subsequent smoothing and optimization operations are ensured.

[0080] The image pixels are weighted and averaged by the convolution Gaussian kernel function, local noise and detail mutations are gradually weakened, the crop boundary and region transition is more smooth and natural, the edge sawtooth and isolated noise caused by classification errors are effectively reduced, the overall coherence and visual consistency of the crop distribution map are improved, and finally a smooth and continuous crop distribution map is obtained, thereby providing a good foundation for subsequent regional consistency enhancement and accurate mapping.

[0081] By analyzing the category similarity of adjacent pixels or regions, using graph cut, conditional random field (CRF) or region growing algorithm, the spatial optimization and boundary adjustment of the classification result are performed, the local neighborhood information is fused to eliminate isolated noise and small cracks, the coherence and consistency of the crop category in the same land block are enhanced, and finally a land block level crop distribution map with clear boundary and stable distribution is generated, thereby improving the accuracy and practical application value of crop recognition.

[0082] The edge extension and Gaussian blur processing not only makes up for the boundary loss and noise interference that may occur in the splicing process, but also improves the spatial coherence and distribution smoothness of the crop category. The regional consistency enhancement further optimizes the category consistency within the land block, eliminates isolated misjudgment and boundary scattering, improves the accuracy and stability of the crop distribution map, ensures the integrity of the remote sensing recognition result in space and the rationality of the actual land block boundary, improves the crop recognition accuracy, adaptability and stability of the multi-band remote sensing image, and provides reliable data support for agricultural management, land use planning and related decision-making.

[0083] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0084] As shown in Figure 4 , it is a functional module diagram of a multi-band remote sensing image crop recognition device provided by an embodiment of the present application.

[0085] In this embodiment of the disclosure, a multi-band remote sensing image crop identification device is provided, which corresponds one-to-one with the multi-band remote sensing image crop identification method described in the above embodiments. For example... Figure 4 As shown, the multi-band remote sensing image crop identification device 100 can be installed in an electronic device. According to its functions, the multi-band remote sensing image crop identification device 100 includes an image data standardization module 101, a pixel data block generation module 102, a pixel data block stitching module 103, a recognition model optimization module 104, a feature image convolution module 105, and a crop distribution map generation module 106. Detailed descriptions of each functional module are as follows: The image data standardization module 101 is used to acquire multi-band remote sensing image data, generate multi-temporal time series data sequences based on the multi-band remote sensing image data, and perform standardization processing on the multi-temporal time series data sequences to obtain standard image data. The pixel data block generation module 102 is used to extract the pixel neighborhood of the standard image data and generate a number of pixel data blocks based on the pixel neighborhood. The pixel data block stitching module 103 is used to stitch the pixel data blocks to obtain a temporal feature image; The recognition model optimization module 104 is used to acquire a standard bar image, randomly mask the standard bar image, and use the masked standard bar image to perform data augmentation on a preset feature extraction model to obtain a crop recognition model. The feature image convolution module 105 is used to acquire crop type information, and to convolve the temporal feature image using the crop type information and the crop recognition model to obtain the crop type in the image. The crop distribution map generation module 106 is used to generate a plot crop distribution map based on the crop types in the image.

[0086] In one embodiment, the image data standardization module 101 acquires multi-band remote sensing image data, generates a multi-temporal time series data sequence based on the multi-band remote sensing image data, and performs standardization processing on the multi-temporal time series data sequence to obtain standard image data, including: The time sequence of the multi-band remote sensing image data is obtained, and the multi-band remote sensing image data is sorted according to the time sequence to obtain a multi-temporal time series data sequence. The multi-phase time series data sequence is radiometrically calibrated to obtain a calibrated time series sequence; The calibrated time series is subjected to atmospheric and geometric corrections to obtain the corrected time series. The cloud shadow coverage area of ​​the corrected time series is marked to obtain a marked time series; The resolution uniform processing is performed on the marking time sequence to obtain standard image data.

[0087] In an embodiment, the pixel data block generation module 102 performs the following steps to extract a pixel neighborhood of the standard image data and generate a plurality of pixel data blocks according to the pixel neighborhood: Randomly selecting a plurality of sample center pixels from the standard image data, and generating a two-dimensional neighborhood window of a preset size according to the sample center pixels one by one; Extracting a pixel neighborhood centered on the sample center pixel on the standard image data according to the size of the two-dimensional neighborhood window; Obtaining all band values in each pixel neighborhood, and stacking the band values in a preset vertical direction to obtain a pixel data block.

[0088] In an embodiment, the pixel data block splicing module 103 performs the following steps to splice the pixel data blocks to obtain a time sequence feature image: Converting the pixel data blocks into a two-dimensional pixel image; Splicing the two-dimensional pixel image in a preset horizontal direction to obtain a spliced pixel image; Performing linear stretching on the spliced pixel image to obtain a stretched pixel image; Performing normalization processing on the stretched pixel image to obtain a time sequence feature image.

[0089] In an embodiment, the recognition model optimization module 104 performs the following steps to randomly mask the standard bar image and use the masked standard bar image to perform data enhancement on a preset feature extraction model to obtain a crop recognition model: Obtaining a masking proportion interval, and masking the standard bar image according to the masking proportion interval to obtain a plurality of initial masking area blocks; Randomly selecting one of the initial masking area blocks as a target area block; Adding Gaussian random noise to the target area block to obtain a sample area image; Optimizing a preset feature extraction model using the sample area image to obtain a crop recognition model.

[0090] In an embodiment, the feature image convolution module 105 performs the following steps to use the crop category information and the crop recognition model to convolve the time sequence feature image to obtain an image crop category: Performing first layer convolution on a band dimension of the time sequence feature image using the crop recognition model to obtain a band spectrum difference; performing second layer convolution on the time dimension of the time sequence feature image to obtain a dynamic change feature; performing third layer convolution on the spatial dimension of the time sequence feature image to obtain a crop growth form; summarizing the waveband spectral difference, the dynamic change feature and the crop growth form into a high-dimensional convolution feature; performing average pooling on the high-dimensional convolution feature to obtain a high-dimensional pooling feature; performing class encoding on the crop category information to obtain a crop category encoding; performing full connection processing on the high-dimensional pooling feature and the crop category encoding to obtain a crop category probability; taking a crop category corresponding to the maximum crop category probability as an image crop category.

[0091] In an embodiment, the crop distribution map generation module 106 generates a field crop distribution map according to the image crop category, including: obtaining a pixel position corresponding to the image crop category, splicing and restoring the image crop category according to the pixel position to obtain an initial crop distribution map; performing edge extension on the initial crop distribution map to obtain an extended crop distribution map; performing Gaussian blur processing on the extended crop distribution map to obtain a smoothed crop distribution map; enhancing the region consistency of the smoothed crop distribution map to obtain a field crop distribution map.

[0092] In the present application, for a multi-band remote sensing image crop recognition device, first, the present application obtains multi-band remote sensing image data, generates multi-temporal time series data sequences according to the multi-band remote sensing image data, and standardizes the multi-temporal time series data sequences to obtain standard image data, which not only significantly improves the quality and comparability of the data, but also provides a stable and reliable data foundation for subsequent crop recognition, change detection, time series analysis and other remote sensing applications. Extract the pixel neighborhood of the standard image data, and generate a plurality of pixel data blocks according to the pixel neighborhood, effectively fuse the local spatial structure and multispectral information, and enhance the data expression ability. Not only does it retain the details of the pixel and the neighborhood correlation, but it also helps to improve the accuracy of subsequent classification and identification. The pixel data blocks are spliced to obtain a time series feature image. By converting the multi-band fused pixel data blocks into a two-dimensional pixel image and splicing along the horizontal direction, combining linear stretching and normalization processing, the spatial and spectral characteristics of multi-temporal and multi-sample can be effectively integrated and standardized, and the contrast and detail performance of the time series feature image are enhanced. Obtain a standard bar image, randomly mask the standard bar image, and use the masked standard bar image to enhance the data of the preset feature extraction model to obtain a crop recognition model. Effectively simulate the occlusion, noise and environmental interference in the remote sensing image, improve the diversity and complexity of the training data. Then, obtain crop type information, use the crop type information and the crop recognition model to convolve the time series feature image to obtain an image crop type. By extracting the spectral difference, dynamic change and spatial form features of the crop in the three dimensions of band, time and space, and fusing them into high-dimensional convolution features for pooling compression, further combining the class encoding of the crop type, and obtaining the probability distribution of each crop type through full connection operation, the class corresponding to the maximum probability is finally taken as the recognition result. Not only does it make full use of the multi-dimensional information of remote sensing data, but also introduces class prior knowledge to improve the model discrimination ability, realizes accurate and efficient recognition of crop types, significantly improves the generalization performance and classification accuracy of the model in complex environments. Finally, generate a plot crop distribution map according to the image crop type to ensure the integrity of the remote sensing recognition result in space and the rationality of the real plot boundary, and improve the precision, adaptability and stability of multi-band remote sensing image crop recognition. The specific limitations of the multi-band remote sensing image crop recognition device can be referred to the limitations of the multi-band remote sensing image crop recognition method in the foregoing, which will not be repeated here. Each module in the above multi-band remote sensing image crop recognition device can be realized by software, hardware and their combinations. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory in the computer device in software form, so that the processor can call and execute the operations of the above modules.

[0093] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used for communication with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a multi-band remote sensing image crop identification method on the server side.

[0094] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a multi-band remote sensing image crop identification method on the client side.

[0095] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Acquire multi-band remote sensing image data, generate multi-temporal time series data sequences based on the multi-band remote sensing image data, and perform standardization processing on the multi-temporal time series data sequences to obtain standard image data; Extract the pixel neighborhood of the standard image data, and generate several pixel data blocks based on the pixel neighborhood; The pixel data blocks are stitched together to obtain a temporal feature image; A standard bar image is obtained, the standard bar image is randomly masked, and the masked standard bar image is used to perform data augmentation on a preset feature extraction model to obtain a crop recognition model. Obtain crop type information, and use the crop type information and the crop recognition model to convolve the temporal feature image to obtain the crop type in the image; A field crop distribution map is generated according to the crop species of the image.

[0096] In several embodiments provided by the present application, it should be understood that the disclosed devices and apparatuses can be implemented in other manners. For example, the above described system embodiments are merely illustrative. For example, the division of the modules is merely logical function division. In actual implementation, other division manners can be adopted.

[0097] In addition, each function module in each embodiment of the present application can be integrated in a processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The above integrated unit can be realized in the form of hardware, or in the form of hardware plus software function module.

[0098] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0099] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0100] In some embodiments of the present embodiment, a computer readable storage medium is provided, and a computer program is stored on the computer readable storage medium, wherein the computer program is executed by a processor to implement the steps of the method described in the above embodiments.

[0101] The readable storage medium of the present application stores a computer program, and the computer program, when executed by a processor of an electronic device, can implement: Obtaining multi-band remote sensing image data, generating a multi-temporal time series data sequence according to the multi-band remote sensing image data, and performing standardization processing on the multi-temporal time series data sequence to obtain standard image data; Extracting a pixel neighborhood of the standard image data, and generating a plurality of pixel data blocks according to the pixel neighborhood; Splicing the pixel data blocks to obtain a time series feature image; Obtaining a standard bar image, randomly masking the standard bar image, and using the masked standard bar image to perform data enhancement on a preset feature extraction model to obtain a crop recognition model; Obtaining crop category information, and performing convolution on the time-series feature image by using the crop category information and the crop recognition model to obtain an image crop category. Generating a plot crop distribution map according to the image crop category.

[0102] It should be noted that the functions or steps described above with respect to the computer-readable storage medium or the computer device can correspond to the related descriptions of the server side and the client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0103] The computer-readable storage medium can also store at least one computer executable program / instruction, such as computer readable instructions. The computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory, etc. The computer-readable storage medium may, for example, include read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer-readable storage medium can be connected to a computing device such as a computer, and then when the computing device runs the computer readable instructions stored on the computer readable storage medium, the various methods described above can be performed.

[0104] In addition, the computer device can also include (but not limited to) a data bus, an input / output (I / O) bus, a display, and an input / output device (for example, a keyboard, a mouse, a speaker, etc.), etc.

[0105] The processor can communicate with external devices through the I / O bus via wired or wireless networks.

[0106] In one embodiment, the at least one computer executable instruction can also be compiled into or constitute a software product / computer program product, wherein one or more computer executable instructions are executed by the processor to perform the steps of the functions and / or methods described in the embodiments of the present technology.

[0107] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0108] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of functional units and modules is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0109] In the embodiments provided by the present disclosure, it should be understood that the disclosed apparatus and method can also be implemented in other manners. The embodiments described above are merely exemplary for describing the present disclosure. For example, the flowcharts and block diagrams in the accompanying drawings show the possible implementation architectures, functions and operation of the apparatus, method and computer program product according to the embodiments of the present disclosure. In this regard, each block in the flowcharts and block diagrams can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing the specified logic function. It should also be noted that, in some alternative implementations, the functions noted in the blocks can occur in different orders from those noted in the accompanying drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and the combination of blocks in the block diagrams and / or flowcharts, can be implemented by a special-purpose hardware-based system for implementing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0110] It should be noted that, in the present disclosure, the term "comprising" or "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without more limitations, the element limited by the statement "including a" does not exclude the presence of additional same elements in the process, method, article or device including the element.

[0111] The above-described embodiments are merely used to illustrate the technical solutions of the present disclosure, rather than limit them; although the present disclosure has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent replacements; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the protection scope of the present disclosure.

[0112] It should be noted that, in the embodiments of the present disclosure, if non-company software tools or components appear, they are only used for example introduction, and do not represent actual use.

Claims

1. A multi-band remote sensing image crop identification method, characterized in that, The method comprises: acquiring multi-band remote sensing image data, generating a multi-temporal time series data sequence according to the multi-band remote sensing image data, and performing standardization processing on the multi-temporal time series data sequence to obtain standard image data; extracting a pixel neighborhood of the standard image data, and generating a plurality of pixel data blocks according to the pixel neighborhood; splicing the pixel data blocks to obtain a time series feature image; acquiring a standard bar image, performing random masking on the standard bar image, and using the masked standard bar image to perform data enhancement on a preset feature extraction model to obtain a crop recognition model; acquiring crop category information, and performing convolution on the time series feature image using the crop category information and the crop recognition model to obtain an image crop category; generating a field crop distribution map according to the image crop category.

2. The multi-band remote sensing image crop identification method of claim 1, wherein, The acquiring multi-band remote sensing image data, generating a multi-temporal time series data sequence according to the multi-band remote sensing image data, and performing standardization processing on the multi-temporal time series data sequence to obtain standard image data comprises: acquiring a time sequence of the multi-band remote sensing image data, sorting the multi-band remote sensing image data according to the time sequence to obtain a multi-temporal time series data sequence; performing radiation calibration on the multi-temporal time series data sequence to obtain a calibrated time series; performing atmospheric correction and geometric correction on the calibrated time series to obtain a corrected time series; labeling a cloud shadow coverage area of the corrected time series to obtain a labeled time series; performing resolution uniform processing on the labeled time series to obtain standard image data.

3. The multi-band remote sensing image crop identification method of claim 1, wherein, The extracting a pixel neighborhood of the standard image data, and generating a plurality of pixel data blocks according to the pixel neighborhood comprises: randomly selecting a plurality of sample center pixel points from the standard image data, and generating a two-dimensional neighborhood window of a preset size one by one according to the sample center pixel points; extracting a pixel neighborhood centered on the sample center pixel point on the standard image data according to the size of the two-dimensional neighborhood window; acquiring all band values in each pixel neighborhood, and stacking the band values in a preset vertical direction to obtain a pixel data block.

4. The multi-band remote sensing image crop identification method of claim 1, wherein, The splicing the pixel data blocks to obtain a time series feature image comprises: converting the pixel data blocks into a two-dimensional pixel image; splicing the two-dimensional pixel image in a preset horizontal direction to obtain a spliced pixel image; performing linear stretching on the spliced pixel image to obtain a stretched pixel image; performing normalization processing on the stretched pixel image to obtain a time series feature image.

5. The multi-band remote sensing image crop identification method of claim 1, wherein, The performing random masking on the standard bar image, and using the masked standard bar image to perform data enhancement on a preset feature extraction model to obtain a crop recognition model comprises: acquiring a masking proportion interval, masking the standard bar image according to the masking proportion interval to obtain a plurality of initial masking area blocks; randomly selecting one of the initial masking area blocks as a target area block; adding Gaussian random noise to the target area block to obtain a sample area image; The sample region image is used to optimize a preset feature extraction model, to obtain a crop recognition model.

6. The multi-band remote sensing image crop identification method of claim 1, wherein, The crop category information and the crop recognition model are used to convolve the time-series feature image, to obtain an image crop category, including: The crop recognition model is used to perform first layer convolution on a wave band dimension of the time-series feature image, to obtain a wave band spectrum difference; A second layer convolution is performed on a time dimension of the time-series feature image, to obtain a dynamic change feature; A third layer convolution is performed on a space dimension of the time-series feature image, to obtain a crop growth form; The wave band spectrum difference, the dynamic change feature and the crop growth form are summarized as high-dimensional convolution features; The high-dimensional convolution features are subjected to average pooling, to obtain high-dimensional pooling features; The crop category information is subjected to class encoding, to obtain crop category encoding; The high-dimensional pooling features and the crop category encoding are subjected to full connection processing, to obtain crop category probabilities; A crop category corresponding to a maximum crop category probability is taken as an image crop category.

7. The multi-band remote sensing image crop identification method of claim 1, wherein, The image crop category is used to generate a field crop distribution map, including: Pixel positions corresponding to the image crop category are obtained, the image crop category is spliced and restored according to the pixel positions, to obtain an initial crop distribution map; Edge extension is performed on the initial crop distribution map, to obtain an extended crop distribution map; Gaussian blur processing is performed on the extended crop distribution map, to obtain a smoothed crop distribution map; The region consistency of the smoothed crop distribution map is enhanced, to obtain a field crop distribution map.

8. A multi-band remote sensing map crop identification device, characterized by, The device includes: An image data standardization module is configured to obtain multi-wave band remote sensing image data, generate a multi-time phase time-series data sequence according to the multi-wave band remote sensing image data, and perform standardization processing on the multi-time phase time-series data sequence, to obtain standard image data; A pixel data block generation module is configured to extract a pixel neighborhood of the standard image data, and generate a plurality of pixel data blocks according to the pixel neighborhood; A pixel data block splicing module is configured to splice the pixel data blocks, to obtain a time-series feature image; An identification model optimization module is configured to obtain a standard bar image, perform random masking on the standard bar image, and perform data enhancement on a preset feature extraction model using the masked standard bar image, to obtain a crop recognition model; A feature image convolution module is configured to obtain crop category information, and convolve the time-series feature image using the crop category information and the crop recognition model, to obtain an image crop category; A crop distribution map generation module is configured to generate a field crop distribution map according to the image crop category.

9. An electronic device, comprising: The electronic device includes: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute a multi-wave band remote sensing image crop recognition method according to any one of claims 1 to 7.

10. A computer readable storage medium storing a computer program, characterized in that, The computer program is executed by a processor to implement a multi-band remote sensing image crop identification method as claimed in any one of claims 1 to 7.

Images