A method of agricultural production supervision based on high-resolution remote sensing images

Abstract

The agricultural production supervision method based on high-resolution remote sensing images provided by the present invention includes: obtaining low-resolution remote sensing images of crop planting areas; preprocessing the low-resolution remote sensing images to obtain high-resolution remote sensing images; extracting features of the high-resolution remote sensing images; and supervising agricultural production based on the features. The method can obtain higher-resolution remote sensing images without changing the original hardware equipment of the system, and complete agricultural production supervision based on the high-resolution remote sensing images, making agricultural production supervision more refined.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing technology, and specifically relates to an agricultural production supervision method based on high-resolution remote sensing images. Background Art

[0002] At present, some agricultural production supervision methods mainly use remote sensing images to achieve supervision, but most of the remote sensing images are low-resolution remote sensing images, and the accuracy of remote sensing images is insufficient, which makes the precision of agricultural production supervision slightly insufficient. Summary of the invention

[0003] In view of the defects in the prior art, the present invention provides an agricultural production supervision method based on high-resolution remote sensing images, which makes agricultural production supervision more refined.

[0004] A method for monitoring agricultural production based on high-resolution remote sensing images, comprising:

[0005] Obtain low-resolution remote sensing images of crop-growing areas;

[0006] Preprocess low-resolution remote sensing images to obtain high-resolution remote sensing images;

[0007] Extract features from high-resolution remote sensing images;

[0008] Regulate agricultural production based on characteristics.

[0009] Furthermore, the preprocessing of low-resolution remote sensing images specifically includes:

[0010] Extracting features of the low-resolution remote sensing image to obtain a first feature map;

[0011] Scaling the first feature map to obtain a second feature map;

[0012] Convolving the second feature map using a depth-wise separable convolution algorithm to obtain a third feature map;

[0013] Input the low-resolution remote sensing image into the bidirectional pyramid structure to obtain the fourth feature map;

[0014] The third feature map and the fourth feature map are added together to obtain a high-resolution remote sensing image.

[0015] Furthermore, the features of low-resolution remote sensing images are extracted specifically including:

[0016] The low-resolution remote sensing image is convolved using multiple dilated convolution algorithms with different dilated rates to obtain multiple dilated convolution results.

[0017] All dilated convolution results are fused to obtain the first feature map.

[0018] Furthermore, scaling the first feature map specifically includes:

[0019] Upsampling is used to scale the first feature map;

[0020] The multi-layer perception mechanism composed of LN and MLP performs 1*1 convolution on the scaled first feature map to obtain the second feature map.

[0021] Furthermore, convolving the second feature map using the depthwise separable convolution algorithm specifically includes:

[0022] Convolve the second feature map using 3*3 and 7*7 depth-separable convolution algorithms to obtain two first depth-wise convolution results;

[0023] Convolve the second feature map using a 5*5 depth-separable convolution algorithm to obtain a second depth-wise convolution result;

[0024] Based on the multi-scale global attention mechanism, the key vector, query vector and value vector of the second deep convolution result at different void rates are calculated;

[0025] Calculate the weights of the two first depth convolution results and the second depth convolution result respectively according to the key vector, the query vector and the value vector;

[0026] Perform 1*1 convolution on the three weights to obtain three shared weights respectively;

[0027] Bring the shared weights into three depth-wise separable convolution algorithms respectively;

[0028] Based on the multi-scale global attention mechanism, the multi-layer perception mechanism composed of LN and MLP is used to fuse the first deep convolution result and the second deep convolution result, and then corrected by 1*1 convolution to obtain the third feature map.

[0029] Furthermore, the bidirectional pyramid structure includes, from top to bottom:

[0030] 3*3 dilated convolution algorithm with a dilation rate of 1, 3*3 dilated convolution algorithm with a dilation rate of 2, 3*3 dilated convolution algorithm with a dilation rate of 3, upsampling processing, 3*3 dilated convolution algorithm with a dilation rate of 3, 3*3 dilated convolution algorithm with a dilation rate of 2, and 3*3 dilated convolution algorithm with a dilation rate of 1.

[0031] Furthermore, the supervision of agricultural production according to characteristics specifically includes:

[0032] Extract spectral and texture features from high-resolution remote sensing images;

[0033] Segment high-resolution remote sensing images according to texture features to obtain multiple sub-images;

[0034] The type and growth stage of crops in each sub-image are identified based on their spectral characteristics.

[0035] Furthermore, segmenting high-resolution remote sensing images based on texture features specifically includes:

[0036] When texture features are repetitive in a local area of ​​a high-resolution remote sensing image, the local area is defined as a regulatory area;

[0037] Obtain historical texture features of the regulatory area;

[0038] When the historical texture features show regular changes over time, the image of the supervision area in the high-resolution remote sensing image is segmented into sub-images.

[0039] Furthermore, the regulation of agricultural production based on characteristics also includes:

[0040] When abnormal spectral features or texture features are detected in the sub-image, it is determined that pests or natural disasters exist in the corresponding supervision area.

[0041] Furthermore, the regulation of agricultural production based on characteristics also includes:

[0042] When the spectral characteristics of the sub-image show soil characteristics, the soil moisture is determined based on the corresponding spectral characteristics;

[0043] Get the types of crops planted in the sub-image history;

[0044] Get the soil moisture range corresponding to the type;

[0045] When the soil moisture does not fall within the soil moisture range, a soil moisture warning is output to remind the user to adjust the soil moisture through manual intervention.

[0046] It can be seen from the above technical scheme that the agricultural production supervision method based on high-resolution remote sensing images provided by the present invention collects low-resolution remote sensing images of crop planting areas, and pre-processes the low-resolution remote sensing images to obtain high-resolution remote sensing images. In this way, higher-resolution remote sensing images can be obtained without changing the original hardware equipment of the system. Finally, the features of the high-resolution remote sensing images are extracted; agricultural production is supervised based on the features. In this way, agricultural production supervision can be completed based on high-resolution remote sensing images, making agricultural production supervision more refined. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0048] Figure 1 A flow chart of an agricultural production supervision method provided as an embodiment.

[0049] Figure 2 A schematic diagram of a high-resolution remote sensing image conversion model provided in an embodiment.

[0050] Figure 3 A flowchart of a method for identifying the type and growth stage of a crop provided in an embodiment.

[0051] Figure 4 A flowchart of an image segmentation method provided in an embodiment.

[0052] Figure 5 A flow chart of a disaster monitoring method provided in an embodiment. DETAILED DESCRIPTION

[0053] The embodiments of the technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and the protection scope of the present invention cannot be limited thereto. It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should be the common meanings understood by those skilled in the art to which the present invention belongs.

[0054] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0055] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0056] As used in this specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0057] Example:

[0058] A method for monitoring agricultural production based on high-resolution remote sensing images, see Figure 1 ,include:

[0059] Obtain low-resolution remote sensing images of crop-growing areas;

[0060] Preprocess low-resolution remote sensing images to obtain high-resolution remote sensing images;

[0061] Extract features from high-resolution remote sensing images;

[0062] Regulate agricultural production based on characteristics.

[0063] In this embodiment, the method collects low-resolution remote sensing images of crop planting areas, pre-processes the low-resolution remote sensing images to obtain high-resolution remote sensing images, so that higher-resolution remote sensing images can be obtained without changing the original hardware equipment of the system. Finally, the features of the high-resolution remote sensing images are extracted; agricultural production is supervised based on the features. In this way, agricultural production supervision can be completed based on high-resolution remote sensing images, making agricultural production supervision more refined.

[0064] Furthermore, in some embodiments, preprocessing the low-resolution remote sensing image specifically includes:

[0065] Extracting features of the low-resolution remote sensing image to obtain a first feature map;

[0066] Scaling the first feature map to obtain a second feature map;

[0067] Convolving the second feature map using a depth-wise separable convolution algorithm to obtain a third feature map;

[0068] Input the low-resolution remote sensing image into the bidirectional pyramid structure to obtain the fourth feature map;

[0069] The third feature map and the fourth feature map are added together to obtain a high-resolution remote sensing image.

[0070] In this embodiment, see Figure 2When converting a low-resolution remote sensing image, this method extracts the features of the low-resolution remote sensing image to obtain a first feature map; scales the first feature map to obtain a second feature map; convolves the second feature map using a depthwise separable convolution algorithm to obtain a third feature map; inputs the low-resolution remote sensing image into a bidirectional pyramid structure to obtain a fourth feature map; and adds the third feature map to the fourth feature map to obtain a high-resolution remote sensing image.

[0071] Furthermore, in some embodiments, extracting features of low-resolution remote sensing images specifically includes:

[0072] The low-resolution remote sensing image is convolved using multiple dilated convolution algorithms with different dilated rates to obtain multiple dilated convolution results.

[0073] All dilated convolution results are fused to obtain the first feature map.

[0074] In this embodiment, see Figure 2 When extracting features of low-resolution remote sensing images, this method uses three branches to convolve the low-resolution remote sensing images, namely 3*3 dilated convolution algorithm (dilated rate = 1) + BN + ReLU, 3*3 dilated convolution algorithm (dilated rate = 2) + BN + ReLU, and 3*3 dilated convolution algorithm (dilated rate = 3) + BN + ReLU, and then fuses the dilated convolution results of the three dilated convolution algorithms through the Efficient Channel Attention (ECA) module to obtain the first feature map. In this way, when extracting features of low-resolution remote sensing images, this method captures the features of low-resolution remote sensing images at different scales through different dilated rates, improves the method's processing ability for multi-scale objects, and uses dilated convolution algorithms with different dilated rates to effectively expand the receptive field, thereby capturing a wider range of contextual information.

[0075] Further, in some embodiments, scaling the first feature map specifically includes:

[0076] Upsampling is used to scale the first feature map;

[0077] The multi-layer perception mechanism composed of LN and MLP performs 1*1 convolution on the scaled first feature map to obtain the second feature map.

[0078] In this embodiment, see Figure 2 ,This method first scales the first feature map by upsampling, and then performs 1*1 convolution on the scaled first feature map based on the multi-layer perception mechanism composed of LN+MLP to obtain the second feature map, and uses the second feature map to construct a high-resolution remote sensing image.

[0079] Further, in some embodiments, convolving the second feature map using a depthwise separable convolution algorithm specifically includes:

[0080] Convolve the second feature map using 3*3 and 7*7 depth-separable convolution algorithms to obtain two first depth-wise convolution results;

[0081] Convolve the second feature map using a 5*5 depth-separable convolution algorithm to obtain a second depth-wise convolution result;

[0082] Based on the multi-scale global attention mechanism, the key vector, query vector and value vector of the second deep convolution result at different void rates are calculated;

[0083] Calculate the weights of the two first depth convolution results and the second depth convolution result respectively according to the key vector, the query vector and the value vector;

[0084] Perform 1*1 convolution on the three weights to obtain three shared weights respectively;

[0085] Bring the shared weights into three depth-wise separable convolution algorithms respectively;

[0086] Based on the multi-scale global attention mechanism, the multi-layer perception mechanism composed of LN and MLP is used to fuse the first deep convolution result and the second deep convolution result, and then corrected by 1*1 convolution to obtain the third feature map.

[0087] In this embodiment, see Figure 2 , this method uses three branches to perform operations when reconstructing high-resolution remote sensing images. The three branches are 3*3 depth-separable convolution algorithm, 5*5 depth-separable convolution algorithm, and 7*7 depth-separable convolution algorithm. After the 5*5 depth-separable convolution algorithm, this method also adds a multi-scale global attention mechanism (Multi-Scale DilateFormer Attention, MSDA) to realize the calculation of different vectors (key vector, query vector and value vector) through different void rates, and then integrates them into the 3*3 depth-separable convolution algorithm and the 7*7 depth-separable convolution algorithm to calculate the weights of the three branches respectively. Then, this method shares the weights of the three branches through 1*1 convolution to obtain three shared weights, and passes them into the 3*3 depth-separable convolution algorithm, the 5*5 depth-separable convolution algorithm, and the 7*7 depth-separable convolution algorithm respectively. Finally, a multi-layer perception mechanism of multi-scale global attention mechanism (MSDA) + LN layer + MLP is embedded, and the number of channels is corrected through 1*1 convolution to obtain the third feature map.

[0088] Furthermore, in some embodiments, the bidirectional pyramid structure includes, from top to bottom:

[0089] 3*3 dilated convolution algorithm with a dilation rate of 1, 3*3 dilated convolution algorithm with a dilation rate of 2, 3*3 dilated convolution algorithm with a dilation rate of 3, upsampling processing, 3*3 dilated convolution algorithm with a dilation rate of 3, 3*3 dilated convolution algorithm with a dilation rate of 2, and 3*3 dilated convolution algorithm with a dilation rate of 1.

[0090] In this embodiment, see Figure 2 This method establishes a long-distance residual module between low-resolution remote sensing images and high-resolution remote sensing images to learn long-distance dependencies. The long-distance residual module adopts a bidirectional pyramid structure. The bidirectional pyramid structure includes, from top to bottom: a 3*3 dilated convolution algorithm with a dilation rate of 1, a 3*3 dilated convolution algorithm with a dilation rate of 2, a 3*3 dilated convolution algorithm with a dilation rate of 3, upsampling processing, a 3*3 dilated convolution algorithm with a dilation rate of 3, a 3*3 dilated convolution algorithm with a dilation rate of 2, and a 3*3 dilated convolution algorithm with a dilation rate of 1.

[0091] Further, in some embodiments, see Figure 3 , the supervision of agricultural production according to its characteristics specifically includes:

[0092] Extract spectral and texture features from high-resolution remote sensing images;

[0093] Segment high-resolution remote sensing images according to texture features to obtain multiple sub-images;

[0094] The type and growth stage of crops in each sub-image are identified based on their spectral characteristics.

[0095] In this embodiment, the method extracts the features in the high-resolution remote sensing image as spectral features and texture features. The method first segments the high-resolution remote sensing image according to the texture features to obtain multiple sub-images. For example, the image is segmented according to the texture features of adjacent pixels. The features in a single sub-image should be kept as consistent as possible, and there are differences in the features between different sub-images. Then, the type and growth stage of the crops in each sub-image are identified according to the spectral features. Generally, different crops have different reflectances to spectra of different bands at different growth stages. For example, in the visible light band, the chlorophyll content of the leaves will affect the reflectance; in the near-infrared band, the cell structure and water content of the leaves will affect the reflectance, so the method can identify the type and growth stage of the crops in each sub-image according to the spectral features.

[0096] Further, in some embodiments, see Figure 4 , segmenting high-resolution remote sensing images based on texture features specifically includes:

[0097] When texture features are repetitive in a local area of ​​a high-resolution remote sensing image, the local area is defined as a regulatory area;

[0098] Obtain historical texture features of the regulatory area;

[0099] When the historical texture features show regular changes over time, the image of the supervision area in the high-resolution remote sensing image is segmented into sub-images.

[0100] In this embodiment, when a large area of ​​crops are planted in a certain area, the texture features of the high-resolution remote sensing image in the area will show a certain repeatability, and the repeated coarseness, smoothness, granularity and other characteristics are determined according to the type and growth stage of the crops. For some areas where crops are not planted (such as light soil), the texture features of the high-resolution remote sensing image may also show a certain repeatability. Therefore, when the texture features are repeatable in a local area of ​​the high-resolution remote sensing image, it means that crops may have been planted in the local area, and further confirmation is required. The further confirmation method is to determine it based on the historical texture features of the area. Since the area where crops are planted changes with the growth stage of the crops, its texture features will show regular changes in time. Therefore, when the historical texture features show regular changes over time, it means that the supervision area is an area where crops are planted, and it is necessary to segment the image of the supervision area in the high-resolution remote sensing image into sub-images, and supervise the growth of crops through the features of the sub-images.

[0101] Furthermore, in some embodiments, supervising agricultural production according to the characteristics further includes:

[0102] When abnormal spectral features or texture features are detected in the sub-image, it is determined that pests or natural disasters exist in the corresponding supervision area.

[0103] In this embodiment, the method can also monitor pests and diseases or natural disasters. Since the spectral reflectance of crops infested by pests and diseases or natural disasters in certain bands is different from that of healthy crops, the reflectance may be reduced or increased, so the method detects pests and diseases or natural disasters by analyzing spectral features. For example, when it is detected that there is an abnormal change in the spectral curve in the sub-image, the color in the sub-image turns yellow or darkens, and the texture features in the sub-image become rough, messy or missing, it is determined that the crops have been affected by pests and diseases. When the hue in the sub-image is different from the surrounding hue, it is determined that there is flooding in the supervision area. When the greenness in the sub-image is reduced or the texture is dry, it is determined that there is a drought in the supervision area. When the sub-image color is blackened, it is determined that there is a fire in the supervision area, etc. When the method detects that there are pests and diseases or natural disasters in the supervision area, it can promptly notify the administrator to take active measures to grasp the area and degree of disaster in the affected area.

[0104] Further, in some embodiments, see Figure 5 , monitoring of agricultural production according to characteristics also includes:

[0105] When the spectral characteristics of the sub-image show soil characteristics, the soil moisture is determined based on the corresponding spectral characteristics;

[0106] Get the types of crops planted in the sub-image history;

[0107] Get the soil moisture range corresponding to the type;

[0108] When the soil moisture does not fall within the soil moisture range, a soil moisture warning is output to remind the user to adjust the soil moisture through manual intervention.

[0109] In this embodiment, since different soil moisture for different crops will directly affect the yield of crops, the method can also monitor soil moisture. When the spectral characteristics of the sub-image show soil characteristics, it means that the crops planted in the area have been harvested at this time, and the soil moisture can be determined according to the corresponding spectral characteristics. In general, the planting areas of most crops are relatively fixed, so the method obtains the types of crops planted in the sub-image history to obtain the types of crops planted in the area in the past. The method sets different soil moisture ranges for different crops. As long as the soil moisture is within the soil moisture range, the impact of soil moisture on crop yield will be minimized. If the soil moisture does not fall within the soil moisture range, a soil moisture warning is output to remind the user to adjust the soil moisture through manual intervention. For example, when the soil moisture is not enough, the soil moisture can be adjusted by artificial rainfall, irrigation, etc., so that the soil moisture is within the soil moisture range.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A method for monitoring agricultural production based on high-resolution remote sensing images, characterized in that: include: Obtain low-resolution remote sensing images of crop-growing areas; Preprocess low-resolution remote sensing images to obtain high-resolution remote sensing images; Extracting features of the high-resolution remote sensing image; Regulation of agricultural production according to the characteristics described; The preprocessing of the low-resolution remote sensing image specifically includes: Extracting features of the low-resolution remote sensing image to obtain a first feature map; Scaling the first feature map to obtain a second feature map; Convolving the second feature map using a depthwise separable convolution algorithm to obtain a third feature map; Inputting the low-resolution remote sensing image into a bidirectional pyramid structure to obtain a fourth feature map; Adding the third feature map and the fourth feature map to obtain the high-resolution remote sensing image; Extracting the features of the low-resolution remote sensing image specifically includes: Convolving the low-resolution remote sensing image using multiple dilated convolution algorithms with different dilated rates to obtain multiple dilated convolution results; All the dilated convolution results are fused through efficient channel attention to obtain the first feature map.

2. The agricultural production supervision method based on high-resolution remote sensing images according to claim 1 is characterized in that: Scaling the first feature map specifically includes: Scaling the first feature map by upsampling; A 1*1 convolution is performed on the scaled first feature map based on a multi-layer perception mechanism composed of LN and MLP to obtain the second feature map.

3. The agricultural production supervision method based on high-resolution remote sensing images according to claim 2 is characterized in that: Convolving the second feature map using a depthwise separable convolution algorithm specifically includes: Convolving the second feature map using a 3*3 and 7*7 depthwise separable convolution algorithm to obtain two first depthwise convolution results; Convolving the second feature map using a 5*5 depthwise separable convolution algorithm to obtain a second depthwise convolution result; Based on the multi-scale global attention mechanism, the key vector, query vector and value vector of the second deep convolution result at different void rates are calculated; Calculate weights of the two first depth convolution results and the second depth convolution result respectively according to the key vector, the query vector and the value vector; Perform 1*1 convolution on the three weights to obtain three shared weights respectively; The shared weights are respectively introduced into three depth-separable convolution algorithms; Based on the multi-scale global attention mechanism, the multi-layer perception mechanism composed of LN and MLP, the first deep convolution result and the second deep convolution result are fused and corrected by 1*1 convolution to obtain the third feature map.

4. The agricultural production supervision method based on high-resolution remote sensing images according to claim 2 is characterized in that: The bidirectional pyramid structure includes, from top to bottom: 3*3 dilated convolution algorithm with a dilation rate of 1, 3*3 dilated convolution algorithm with a dilation rate of 2, 3*3 dilated convolution algorithm with a dilation rate of 3, upsampling processing, 3*3 dilated convolution algorithm with a dilation rate of 3, 3*3 dilated convolution algorithm with a dilation rate of 2, and 3*3 dilated convolution algorithm with a dilation rate of 1.

5. The agricultural production supervision method based on high-resolution remote sensing images according to claim 1 is characterized in that: The supervision of agricultural production according to the characteristics specifically includes: Extracting spectral features and texture features from the high-resolution remote sensing image; Segmenting the high-resolution remote sensing image according to texture features to obtain a plurality of sub-images; The type and growth stage of crops in each sub-image are identified based on their spectral characteristics.

6. The agricultural production supervision method based on high-resolution remote sensing images according to claim 5 is characterized in that: The segmentation of the high-resolution remote sensing image according to the texture features specifically includes: When the texture feature is repetitive in a local area of ​​the high-resolution remote sensing image, the local area is defined as a supervision area; Obtaining historical texture features of the regulatory area; When the historical texture features show regular changes over time, the image of the supervision area in the high-resolution remote sensing image is segmented into sub-images.

7. The agricultural production supervision method based on high-resolution remote sensing images according to claim 6 is characterized in that: The supervision of agricultural production according to the characteristics also includes: When abnormal spectral features or texture features are detected in the sub-image, it is determined that pests or natural disasters exist in the corresponding supervision area.

8. The agricultural production supervision method based on high-resolution remote sensing images according to claim 7 is characterized in that: The supervision of agricultural production according to the characteristics also includes: When the spectral characteristics of the sub-image show soil characteristics, determining soil moisture according to the corresponding spectral characteristics; Obtaining the types of crops planted in the sub-image historically; Get the soil moisture range corresponding to the type; When the soil moisture does not fall within the soil moisture range, a soil moisture warning is output to remind the user to adjust the soil moisture through manual intervention.

Images