A method for eliminating interference information of crop canopy temperature

Abstract

The present invention relates to a method for removing interference information of crop canopy temperature, and the main steps are as follows: obtaining multi-spectral images, thermal images and RGB images of farmland; if the time interval between obtaining the multi-spectral image and the thermal image of the farmland is less than 30 minutes, first, using a segmentation algorithm for the multi-spectral image to automatically segment the vegetation index image to generate a binary image, and removing bare soil and shadows therefrom; using a segmentation algorithm for the thermal image to automatically segment the temperature image of the farmland to generate a binary image, and removing the ridge and its thermal effect area therefrom; spatially superimposing the above two binary images to generate a crop canopy temperature image of the farmland; if the time interval between obtaining the multi-spectral image and the thermal image of the farmland is greater than 30 minutes, using the RGB image obtained synchronously with the thermal image to replace the multi-spectral image. This method can quickly extract the crop canopy temperature and improve the accuracy of canopy temperature information.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural production, and specifically, to a method for eliminating interference information of crop canopy temperature. Background Art

[0002] In the fields of rapid screening of stress-resistant crop varieties, evaluation of water and fertilizer status, and yield prediction, the comprehensive application of crop canopy spectrum, texture, temperature, and structure is the focus of agricultural image research. As one of the important characteristic parameters, temperature can be obtained through the conversion of UAV thermal images. Generally, the temperature image contains information such as crops, bare soil, and weeds. Weeds are less common due to the use of pesticides, so they can be ignored. The bare soil mainly comes from the ridges and the gap areas where crop plants grow. Under sunlight, the upper part of the plant forms a shadow due to projection onto the canopy, resulting in a significant decrease in temperature in the shadow area. In addition, the soil surface of the ridge not only has a significantly higher temperature than the crop canopy, but also the thermal radiation of the soil surface causes a significant increase in the temperature of the crop canopy in the area adjacent to the soil. Therefore, the information interfering with the accuracy of canopy temperature mainly includes three categories, namely: soil, shadow, and the area where the temperature of the surrounding crop canopy is abnormally increased due to the thermal radiation of the ridge (thermal effect area). To achieve the purpose of rapid extraction of canopy temperature, an automatic elimination method for the above three types of interference information must be constructed. Up to now, although many dynamic threshold segmentation algorithms have been developed, their application scenarios are different and their applicability is also different. Therefore, the optimal segmentation method must be found from them. In addition, simply using the temperature image cannot effectively segment all interference information.

[0003] The ground object classification methods based on multi - spectral images can be roughly divided into three categories, namely the manual threshold method of vegetation index, the supervised image classification method, and the automatic threshold method of vegetation index. The manual threshold method of vegetation index uses the normalized difference vegetation index (NDVI), soil - adjusted vegetation index (SAVI), or normalized difference canopy shadow index (NDCSI), etc. By comparing the pixel value differences of crops, bare soil, and shaded pixels on the image, a threshold is set manually for differentiation. The supervised image classification method must first mark a certain number of crop, bare soil, and shadow samples, and use spectral, texture, and spatial information, etc. as sample classification parameters. Then, methods such as principal component analysis, K - nearest neighbor method, and support vector machine are used to automatically identify and classify within the entire field. The automatic threshold method of vegetation index refers to the dynamic threshold segmentation method, including the Otsu method, Iterative Self - Organizing Data Analysis (ISODATA), Maximum Entropy method (MAXENTROPY), Mean method (MEAN), Minimum Error method (MINERROR), and Moments method (MOMENTS), etc. These methods can automatically determine the vegetation index thresholds between different categories, and then distinguish crops, bare soil, and shadows according to this threshold. The manual threshold method of vegetation index must compare the image segmentation accuracy by setting different thresholds manually, and then determine the optimal threshold. When processing fields with different growth conditions on the image, the optimal threshold must be searched for each field, which is time - consuming and laborious and is not suitable for the purpose of automatically eliminating the interference information of canopy temperature. The supervised image classification method has a relatively high overall accuracy, but manual participation is required in the early stage to train classification samples, and it is not a completely automatic classification process, so it is also not suitable for the purpose of automatically eliminating the interference information of canopy temperature. The automatic threshold method of vegetation index is based on the principle of automatic segmentation of dynamic thresholds and does not require manual participation throughout the process, so it is used as the technical basis of the present invention.

[0004] When the multi - spectral image and the thermal image of the unmanned aerial vehicle (UAV) come from different UAV platforms, due to the different acquisition times of the two, affected by the changes in the solar altitude angle and azimuth angle, the positions and shapes of the shadows on the multi - spectral and thermal images will also change accordingly. Therefore, the segmentation result based on the multi - spectral image is no longer applicable to the thermal image. At this time, the RGB image acquired by the same UAV as the thermal image is used to segment crops, bare soil, and shadows. The method is similar to that of the multi - spectral image and is also divided into three categories, namely the manual threshold method of vegetation index, the RGB image supervised classification method, and the automatic threshold method of vegetation index. However, the vegetation indices used are different, and common indices such as the Excess Green index (ExG), Excess Green minus Excess Red index (ExGR), and Green - Red Vegetation Index (GRVI), etc. Current research mainly focuses on the threshold segmentation of UAV images for bare soil and shadows, and there is less research on the method of eliminating the area where the canopy temperature of surrounding crops is abnormally increased caused by the thermal radiation of the ridge soil surface, resulting in a decrease in the accuracy of extracting the canopy temperature of crops. Summary of the Invention

[0005] The present invention proposes to use the temperature anomaly caused by the thermal radiation on the surface of the ridge as another major noise source, constructs a method for automatically removing interference information such as soil, shadow, and ridge thermal radiation effects, and proposes a method for screening the optimal vegetation index and segmentation algorithm, thereby effectively improving the accuracy of extracting the crop canopy temperature.

[0006] The content of the present invention is as follows:

[0007] The present invention proposes a method for removing interference information of crop canopy temperature, including the following steps:

[0008] Obtain the multi-spectral image, thermal image, and RGB image of the farmland;

[0009] If the time interval between obtaining the multi-spectral image and the thermal image of the farmland is less than 30 minutes, then first use the dynamic threshold segmentation algorithm for the multi-spectral image to automatically segment the vegetation index image of the farmland to generate a binary image, and remove the bare soil and shadow from it; use the dynamic threshold segmentation algorithm for the thermal image to automatically segment the temperature image of the farmland to generate a binary image, and remove the ridge and its thermal effect area from it; spatially superimpose the above two binary images to generate the crop canopy temperature image of the farmland;

[0010] If the time interval between obtaining the multi-spectral image and the thermal image of the farmland is greater than 30 minutes, then use the RGB image obtained synchronously with the thermal image to replace the multi-spectral image. First, use the dynamic threshold segmentation algorithm for the RGB image to automatically segment the vegetation index image of the farmland to generate a binary image, and remove the bare soil and shadow from it; use the dynamic threshold segmentation algorithm for the thermal image to automatically segment the temperature image of the farmland to generate a binary image, and remove the ridge and its thermal effect area from it; spatially superimpose the above two binary images to generate the crop canopy temperature image of the farmland.

[0011] Further, after obtaining the multi-spectral image, thermal image, and RGB image of the farmland, the following steps are further included:

[0012] Screen the optimal vegetation index and dynamic threshold segmentation algorithm based on the multi-spectral image of the farmland, including the following steps: Cut out an image on the multi-spectral image of the farmland that simultaneously includes crop, ridge and other ground object information; Use image segmentation technology to perform multi-scale segmentation on the multi-spectral image of the farmland to generate different block patches, that is, objects; Randomly extract a certain number of the objects, and visually interpret them into three categories: crops, soil, and shadow as classification training samples; Under the support of the supervised classification method, use the training samples to identify unknown objects and complete the classification of all objects in the farmland; Use different dynamic threshold segmentation algorithms to automatically segment the vegetation index image calculated based on the multi-spectral image reflectance, and screen out the optimal vegetation index and the optimal dynamic threshold segmentation algorithm according to the segmentation accuracy;

[0013] Screen the optimal vegetation indices and segmentation algorithms based on farmland RGB images, including the following steps: Crop an image on the farmland RGB image that includes the information of ground objects such as crops and ridges; Use image segmentation technology to perform multi-scale segmentation and supervised classification on the RGB image of the farmland; Use different dynamic threshold segmentation algorithms to automatically segment the vegetation index image based on the chromaticity coordinates of the RGB image, and screen out the optimal vegetation index and the optimal dynamic threshold segmentation algorithm according to the segmentation accuracy;

[0014] Screen the optimal segmentation algorithm based on farmland thermal images, including the following steps: Convert the pixel values of the farmland thermal image into Celsius temperature to generate a temperature image; Crop an image on the temperature image that includes the information of ground objects such as crops and ridges; Use different dynamic threshold segmentation algorithms to automatically segment the temperature image to generate a binary image of the segmentation result, and screen out the optimal segmentation algorithm based on the thermal image according to the segmentation accuracy.

[0015] Furthermore, the vegetation indices calculated based on the reflectance of multi-spectral images include the Normalized Difference Vegetation Index (NDVI), the Soil Adjusted Vegetation Index (SAVI), and the Normalized Difference Canopy Shadow Index (NDCSI):

[0016]

[0017] Among them, R NIR and R NIR represent the reflectances of the near-infrared and red bands respectively;

[0018]

[0019] Among them, L represents the soil adjustment factor, usually taking 0.5;

[0020]

[0021] Among them, R RE represents the reflectance of the red edge band, R RE_max and R RE_min represent the maximum and minimum red edge reflectances of all pixels in the image respectively. In practical applications, they can be determined by the thresholds at specific ranges (such as the 1% and 99% positions) on the histogram.

[0022] Furthermore, the dynamic threshold segmentation algorithms include the Otsu method (OTSU), the Iterative Self-Organizing Data Analysis method (ISODATA), the Maximum Entropy method (MAXENTROPY), the Mean method (MEAN), the Minimum Error method (MINERROR), and the Moments method (MOMENTS).

[0023] Furthermore, the vegetation indices based on the chromaticity coordinates of RGB images include the Excess Green Index (ExG), the Excess Green minus Excess Red Index (ExGR), and the Green-Red Vegetation Index (GRVI):

[0024] ExG = 2g - r - b

[0025] ExR = 1.4r - g

[0026] ExGR = ExG - ExR

[0027]

[0028] where r, g, and b are the chromaticity coordinates respectively:

[0029]

[0030]

[0031]

[0032] where R, G, and B are the actual pixel values on the RGB image.

[0033] Furthermore, when the crop is winter wheat, the optimal vegetation index based on the multispectral image is the Normalized Difference Canopy Shadow Index (NDCSI), and the optimal dynamic threshold segmentation algorithm is the Otsu method (OTSU).

[0034] Furthermore, when the crop is winter wheat, the optimal vegetation index based on the RGB image is the Excess Green minus Excess Red Index (ExGR), and the optimal dynamic threshold segmentation algorithm is the Minimum Error Threshold method (MINERROR).

[0035] Furthermore, when the crop is winter wheat, the optimal segmentation algorithm based on the thermal image is the Histogram Threshold method (MEAN).

[0036] The present invention proposes a method for removing the interference information of crop canopy temperature, and constructs a method for screening the optimal vegetation index and segmentation algorithm. The crops are segmented from bare soil and shadows using multispectral or RGB images, and the crops are segmented from the areas with abnormally elevated canopy temperature caused by the thermal radiation of the ridge soil surface using the temperature image. Then, the two segmented images are spatially superimposed to achieve the purpose of removing interference information and improving the extraction accuracy of crop canopy temperature. Description of the Drawings

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0038] Figure 1 Flowchart of the method for removing interference information of crop canopy temperature in the embodiment of the present invention;

[0039] Figure 2 Schematic diagram of the technical route of the method for removing interference information of crop canopy temperature in the embodiment of the present invention;

[0040] Figure Comparison of the average temperature of the ridge and its thermal effect area generated based on 6 dynamic threshold segmentation algorithms, and different distance buffers on April 28 (A), May 12 (B), and May 21 (C), 2021. Detailed implementation manners

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0042] According to an embodiment of the present invention, a method for removing interference information of crop canopy temperature is provided. Figure 1 The flowchart of the method for removing interference information of crop canopy temperature includes the following steps:

[0043] Obtain the multi-spectral image, thermal image, and RGB image of the farmland; if the time interval between obtaining the multi-spectral image and the thermal image of the farmland is less than 30 minutes, first use the segmentation algorithm for the multi-spectral image to automatically segment the vegetation index image of the farmland to generate a binary image, and remove the bare soil and shadows from it; use the segmentation algorithm for the thermal image to automatically segment the temperature image of the farmland to generate a binary image, and remove the ridge and its thermal effect area from it; spatially superimpose the above two binary images to generate the crop canopy temperature image of the farmland.

[0044] If the time interval between the acquisition of the farmland multispectral image and the thermal image is greater than 30 minutes, use the RGB image acquired synchronously with the thermal image to replace the multispectral image. First, use the segmentation algorithm for RGB images to automatically segment the vegetation index image of the farmland, generate a binary image, and remove bare soil and shadows therefrom; use the segmentation algorithm for thermal images to automatically segment the temperature image of the farmland, generate a binary image, and remove the ridge and its thermal effect area therefrom; spatially superimpose the above two binary images to generate a farmland canopy temperature image.

[0045] Through the above method, the crops are segmented from the bare soil and shadows using the multispectral or RGB image, and the crops are segmented from the area where the canopy temperature abnormally increases caused by the thermal radiation of the ridge soil surface using the temperature image, and then the two segmented images are spatially superimposed to improve the accuracy of the canopy temperature information.

[0046] In a preferred embodiment of the present invention, to improve the accuracy of the acquired crop canopy temperature, the optimal vegetation index and the optimal dynamic threshold segmentation algorithm based on the farmland multispectral image and the farmland RGB image, and the optimal segmentation algorithm based on the farmland thermal image are screened. That is, after the acquisition of the farmland multispectral image, thermal image and RGB image, the following steps are added:

[0047] Screen the optimal vegetation index and the dynamic threshold segmentation algorithm based on the farmland multispectral image, including the following steps: Cut out an image on the farmland multispectral image that includes ground object information such as crops and ridges; use the image segmentation technology to perform multi-scale segmentation on the farmland multispectral image to generate different block patches, that is, objects; randomly select a certain number of the objects and divide them into three categories: crops, soil, and shadows through visual interpretation as classification training samples; under the support of the supervised classification method, use the training samples to identify unknown objects and complete the classification of all objects in the farmland; use different dynamic threshold segmentation algorithms to automatically segment the vegetation index image calculated based on the reflectance of the multispectral image, and screen out the optimal vegetation index and the optimal dynamic threshold segmentation algorithm according to the segmentation accuracy;

[0048] Screen the optimal vegetation index and the segmentation algorithm based on the farmland RGB image, including the following steps: Cut out an image on the farmland RGB image that includes ground object information such as crops and ridges; use the image segmentation technology to perform multi-scale segmentation and supervised classification on the farmland RGB image; use different dynamic threshold segmentation algorithms to automatically segment the vegetation index image based on the chromaticity coordinates of the RGB image, and screen out the optimal vegetation index and the optimal dynamic threshold segmentation algorithm according to the segmentation accuracy;

[0049] Screen the optimal segmentation algorithm based on farmland thermal images, including the following steps: convert the pixel values of the farmland thermal image into Celsius temperature to generate a temperature image; crop an image on the temperature image that includes ground object information such as crops and ridges; use different dynamic threshold segmentation algorithms to automatically segment the temperature image to generate a binary image of the segmentation result, and screen the optimal segmentation algorithm based on the thermal image according to the segmentation accuracy.

[0050] In a preferred embodiment of the present invention, the vegetation indices calculated based on the reflectance of the multispectral image include the Normalized Difference Vegetation Index (NDVI), the Soil Adjusted Vegetation Index (SAVI), and the Normalized Difference Canopy Shadow Index (NDCSI):

[0051]

[0052] Among them, R NIR and R NIR represent the reflectances of the near-infrared and red bands respectively;

[0053]

[0054] Among them, L represents the soil adjustment factor, usually taken as 0.5;

[0055]

[0056] Among them, R RE represents the reflectance of the red edge band, R RE_max and R RE_min represent the maximum and minimum red edge reflectances of all pixels in the image respectively. In practical applications, they can be determined by the thresholds at specific ranges (such as the 1% and 99% positions) on the histogram.

[0057] The dynamic threshold segmentation algorithms include the Otsu method, the Iterative Self-Organizing Data Analysis Algorithm (ISODATA), the Maximum Entropy method (MAXENTROPY), the Mean method, the Minimum Error Threshold method (MINERROR), and the Moments method.

[0058] The vegetation indices based on the chromaticity coordinates of the RGB image include the Excess Green Index (ExG), the Excess Green minus Excess Red Index (ExGR), and the Green-Red Vegetation Index (GRVI):

[0059] ExG = 2g - r - b

[0060] ExR = 1.4r - g

[0061] ExGR = ExG - ExR

[0062]

[0063] where r, g, and b are chromaticity coordinates respectively:

[0064]

[0065]

[0066]

[0067] where R, G, and B are the actual pixel values on the RGB image.

[0068] In a preferred embodiment of the present invention, when the crop is winter wheat, the optimal vegetation index based on the multispectral image is the Normalized Difference Canopy Shadow Index (NDCSI), and the optimal dynamic threshold segmentation algorithm is the Otsu method; the optimal vegetation index based on the RGB image is the Excess Green minus Excess Red Index (ExGR), and the optimal dynamic threshold segmentation algorithm is the Minimum Error Threshold method (MINERROR); the optimal segmentation algorithm based on the thermal image is the Histogram Threshold method (MEAN).

[0069] Taking the unmanned aerial vehicle flight measurement experiment carried out in the winter wheat field in Hengshui area, Hebei Province on April 28, May 12, and May 21, 2021 as an example, the application of the method for removing the interference information of the crop canopy temperature proposed by the present invention is specifically described. The three dates respectively represent the heading, flowering, and filling stages of winter wheat. Figure 2 It is a schematic technical route diagram of the method for removing the interference information of the crop canopy temperature.

[0070] The multispectral image is obtained by using the DJI Phantom 4 Multispectral Edition, the thermal image is obtained by using the Zenmuse XT2 dual-band thermal imaging camera carried on the DJI Matrice 200 unmanned aerial vehicle, and the RGB image is obtained by using the RGB camera carried on the same aircraft as the thermal imaging camera.

[0071] 1. Screening of the optimal vegetation index and segmentation algorithm based on the unmanned aerial vehicle multispectral image, the steps are as follows:

[0072] (1) Cut out a small piece of the field on the multispectral image that includes the ground object information such as winter wheat and field ridges.

[0073] (2) Object-oriented supervised classification: Use the image segmentation technology to perform multi-scale segmentation on the multispectral image of the small piece of the field to generate different block patches, that is, objects; randomly select a certain number of objects and divide them into three categories: winter wheat, soil, and shadow through visual interpretation as classification training samples; under the support of the supervised classification method, use the training samples to identify unknown objects and complete the classification of all objects in the small field block.

[0074] (3) The vegetation indices based on the reflectance of the multispectral images are the Normalized Difference Vegetation Index (NDVI), the Soil-Adjusted Vegetation Index (SAVI), and the Normalized Difference Canopy Shadow Index (NDCSI).

[0075] (4) Candidates for the dynamic threshold segmentation algorithm: Here, the Otsu method (OTSU), the Iterative Self-Organizing Data Analysis method (ISODATA), the Maximum Entropy method (MAXENTROPY), the Mean method (MEAN), the Minimum Error method (MINERROR), and the Moments method (MOMENTS) are selected as six candidate algorithms.

[0076] (5) Determination of the optimal vegetation index and segmentation algorithm: Using the above six dynamic threshold segmentation algorithms, the above three vegetation index images calculated based on the multispectral image reflectance are automatically segmented respectively, and their segmentation accuracies are evaluated based on the object-oriented classification results in step (2). The evaluation method uses the harmonic mean (F1-score) index of precision and recall. The evaluation accuracies are shown in Table 1. The results show that on May 12th and 21st, 2021, the OTSU algorithm has the highest accuracy in automatically segmenting the NDCSI images, reaching 96.6% and 96.1% respectively; on April 28th, the OTSU and MOMENTS algorithms have the highest accuracy in automatically segmenting the SAVI images, both being 97%. Generally speaking, the OTSU algorithm has the highest average segmentation accuracy for the NDCSI images in each period, reaching 96.5%, which is the best among all the automatic segmentation algorithms. Therefore, NDCSI and OTSU are respectively determined as the optimal vegetation index and segmentation algorithm for the UAV multispectral images.

[0077] Table 1 Automatic segmentation accuracies of three vegetation indices (NDCSI, SAVI, and NDVI) images evaluated based on the F1-score index (unit: %)

[0078]

[0079] 2. Screening of the optimal vegetation index and segmentation algorithm based on UAV RGB images

[0080] The steps are as follows.

[0081] (1) Cut out a small piece of the field on the RGB image that includes the information of winter wheat, ridges and other ground objects.

[0082] (2) Object-oriented supervised classification: Use the image segmentation technology to perform multi-scale segmentation and supervised classification on the RGB image of this small piece of the field. The specific method is similar to that of the multispectral image.

[0083] (3) The vegetation indices based on the chromaticity coordinates of RGB images are the Excess Green Index (ExG), the Excess Green minus Excess Red Index (ExGR), and the Green-Red Vegetation Index (GRVI).

[0084] (4) Candidates for dynamic threshold segmentation algorithms: Similar to the multispectral images, here the Otsu method (OTSU), the Iterative Self-Organizing Data Analysis method (ISODATA), the Maximum Entropy method (MAXENTROPY), the Mean method (MEAN), the Minimum Error method (MINERROR), and the Moments method (MOMENTS) are selected as six candidate algorithms.

[0085] (5) Determination of the optimal vegetation index and segmentation algorithm: Using the above six dynamic threshold segmentation algorithms, the above three vegetation index images calculated based on the chromaticity coordinates of RGB images are automatically segmented respectively, and their segmentation accuracies are evaluated based on the object-oriented classification results in step (2). The evaluation method uses the harmonic mean (F1-score) index of precision and recall. The evaluation accuracies are shown in Table 2. The results show that on April 28, 2021, the MINERROR algorithm has the highest automatic segmentation accuracy for the ExG image, reaching 95.5%; on May 12, the MINERROR algorithm has the highest automatic segmentation accuracies for the ExGR and GRVI images, both being 95.9%; on May 21, the OTSU algorithm has the highest automatic segmentation accuracy for the ExGR image, being 94.9%. Generally speaking, the MINERROR algorithm has the highest average segmentation accuracy for the ExGR images in each period, reaching 95%, which is the best among all automatic segmentation algorithms. Therefore, ExGR and MINERROR are respectively determined as the optimal vegetation index and segmentation algorithm for UAV RGB images.

[0086] Table 2 Automatic segmentation accuracies of three vegetation indices (ExG, ExGR, and GRVI) images evaluated based on the F1-score index (unit: %)

[0087]

[0088] 3. Screening of the optimal segmentation algorithm based on UAV thermal images, the steps are as follows.

[0089] (1) Convert the pixel values of the thermal image into Celsius temperature to generate a temperature image; cut out a small piece of the field that includes the information of winter wheat, field ridges and other ground objects on the temperature image.

[0090] (2) Candidates for dynamic threshold segmentation algorithms: Similar to multi-spectral and RGB images, here, the Otsu method, Iterative Self-Organizing Data Analysis (ISODATA), Maximum Entropy method (MAXENTROPY), Mean method (MEAN), Minimum Error method (MINERROR), and Moments method (MOMENTS) are selected as six candidate algorithms.

[0091] (3) Determination of the optimal segmentation algorithm: Using the above six dynamic threshold segmentation algorithms, the above temperature images are automatically segmented respectively to generate binary images of the segmentation results, and the buffer analysis method is used to evaluate the accuracy of the segmentation results. The specific implementation steps of the buffer analysis method are as follows: 1) Convert the binary images of the above six segmentation results into vector maps, and distinguish winter wheat, ridges, and their thermal effect areas. 2) Based on the ridges and their thermal effect areas, expand outward by a certain distance to construct buffers; the principle for determining the maximum distance is to penetrate into the winter wheat area; here, expand outward by distances of 10 cm, 20 cm, 30 cm, and 40 cm respectively to construct four buffers. 3) Use spatial overlay and spatial extraction techniques to regenerate four adjacent polygon areas with a width of 10 cm each. 4) Use spatial zonal statistics techniques to calculate and output the average temperature of all pixels in the ridges and their thermal effect areas, and each polygon buffer area.

[0092] Compare the average temperature in the ridges and their thermal effect areas, and each adjacent polygon buffer area. The results are as Figure 3As shown in the figure. The unfilled columnar boxes in the figure represent the ridges and their thermal effect regions extracted based on the segmentation results of 6 segmentation algorithms. It can be seen that the temperature of the ridges and their thermal effect regions is significantly higher than that of the buffer regions at different distances from them. Comparing the different-distance buffer zones generated by the ISODATA, MEAN, and MINERROR segmentation algorithms, it shows that the average temperature of the 0-10 cm buffer zone is slightly higher than that of the 10-20 cm, 20-30 cm, and 30-40 cm buffer zones. The maximum difference in the average temperature between the buffer zones was 0.66 °C, 0.62 °C, and 0.62 °C respectively on April 28, 2021, 0.9 °C, 0.83 °C, and 0.9 °C respectively on May 12, and 0.84 °C, 0.82 °C, and 1.2 °C respectively on May 21. Comparing the different-distance buffer zones generated by the MAXENTROPY, MOMENTS, and OTSU segmentation algorithms, it shows that the average temperature gradually decreases as the buffer distance increases. The maximum difference in the average temperature between the buffer zones was 1.78 °C, 4.37 °C, and 4.47 °C respectively on April 28, 2021, 3 °C, 5.32 °C, and 4.89 °C respectively on May 12, and 1.7 °C, 2.77 °C, and 2.65 °C respectively on May 21. Comparing these 6 segmentation algorithms, it shows that the ridges and their thermal effect regions generated by the MEAN algorithm basically cover the regions with abnormally increased temperature caused by soil surface thermal radiation, and can effectively reduce its impact on the extraction accuracy of the winter wheat canopy temperature. Therefore, the MEAN algorithm was determined as the optimal segmentation algorithm for UAV thermal images.

[0093] 4. Automatic elimination of interference information of crop canopy temperature based on UAV image combination, the steps are as follows.

[0094] (1) If the time interval between the acquisition of the multispectral image and the thermal image is less than 30 minutes, then first use the optimal segmentation algorithm OTSU for UAV multispectral images to automatically segment the NDCSI images of all winter wheat fields to generate binary images, and eliminate bare soil and shadows from them; use the optimal segmentation algorithm MEAN for UAV thermal images to automatically segment the temperature images of all winter wheat fields to generate binary images, and eliminate the ridges and their thermal effect regions from them; spatially superimpose the above two binary images to generate the winter wheat canopy temperature image.

[0095] (2) If the time interval between the acquisition of the multispectral image and the thermal image is greater than 30 minutes, consider using the RGB image acquired synchronously with the thermal image to replace the multispectral image. First, use the optimal segmentation algorithm MINERROR for UAV RGB images to automatically segment the ExGR images of all winter wheat fields to generate binary images, and remove bare soil and shadows from them; use the optimal segmentation algorithm MEAN for UAV thermal images to automatically segment the temperature images of all winter wheat fields to generate binary images, and remove ridges and their thermal effect areas from them; spatially superimpose the above two binary images to generate the winter wheat canopy temperature image.

[0096] (3) When there are significant differences in the canopy reflectance of winter wheat fields among different stress treatments, the segmentation idea of using local thresholds instead of global thresholds must also be adopted. Before performing threshold segmentation, first partition according to all treatments, and then for each treatment area, perform segmentation operations according to step (1) or (2), and splice the binary images generated for each treatment area.

[0097] Through the above specific examples, it is shown that the dynamic threshold segmentation algorithm screening method proposed by the present invention for farmland multispectral images, RGB images, and thermal images can screen out the optimal vegetation index and threshold segmentation algorithm suitable for farmland multispectral images and RGB images, and the optimal threshold segmentation algorithm suitable for thermal images.

[0098] In previous literature, bare soil and shadows in the field were regarded as noise sources, and the segmentation results of multispectral images or RGB images were spatially superimposed with thermal images to achieve the purpose of removing bare soil and shadows. The present invention proposes to regard the abnormal increase in temperature caused by the thermal radiation of the ridge soil surface as another main noise source, and systematically expounds the ideas and methods for solving the above three noise problems, thereby effectively improving the accuracy of the winter wheat canopy temperature.

[0099] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for removing interference information of crop canopy temperature, characterized in that Including the following steps: Obtain the multi-spectral image, thermal image, and RGB image of the farmland; If the time interval between the acquisition of the multi-spectral image and the thermal image of the farmland is less than 30 minutes, first use the dynamic threshold segmentation algorithm for the multi-spectral image to automatically segment the vegetation index image calculated based on the reflectance of the multi-spectral image to generate a binary image, and remove bare soil and shadows from it; use the dynamic threshold segmentation algorithm for the thermal image to automatically segment the temperature image of the farmland to generate a binary image, and remove the ridge and its thermal effect area from it; spatially superimpose the above two binary images to generate the farmland canopy temperature image; If the time interval between the acquisition of the multi-spectral image and the thermal image of the farmland is greater than 30 minutes, use the RGB image synchronously acquired with the thermal image to replace the multi-spectral image, and use the dynamic threshold segmentation algorithm for the RGB image to automatically segment the vegetation index image based on the chromaticity coordinates of the RGB image to generate a binary image, and remove bare soil and shadows from it; use the dynamic threshold segmentation algorithm for the thermal image to automatically segment the temperature image of the farmland to generate a binary image, and remove the ridge and its thermal effect area from it; spatially superimpose the above two binary images to generate the farmland canopy temperature image.

2. The method for eliminating crop canopy temperature interference information according to claim 1, characterized in that After obtaining the multi-spectral image, thermal image, and RGB image of the farmland, the following steps are further included: Screen the optimal vegetation index and dynamic threshold segmentation algorithm based on the multi-spectral image of the farmland, including the following steps: Cut out an image on the multi-spectral image of the farmland that includes crop and ridge feature information at the same time; Use image segmentation technology to perform multi-scale segmentation on the multi-spectral image of the farmland to generate different block patches, that is, objects; Randomly select a certain number of the objects and visually interpret them into three categories: crops, soil, and shadows as classification training samples; Under the support of the supervised classification method, use the training samples to identify unknown objects and complete the classification of all objects in the farmland; Use different dynamic threshold segmentation algorithms to automatically segment the vegetation index image calculated based on the reflectance of the multi-spectral image, and select the optimal vegetation index and the optimal dynamic threshold segmentation algorithm according to the segmentation accuracy.

3. The method for eliminating crop canopy temperature interference information according to claim 1, wherein After obtaining the multi-spectral image, thermal image, and RGB image of the farmland, the following steps are further included: Screen the optimal vegetation index and segmentation algorithm based on the RGB image of the farmland, including the following steps: Cut out an image on the RGB image of the farmland that includes crop and ridge feature information at the same time; Use image segmentation technology to perform multi-scale segmentation and supervised classification on the RGB image of the farmland; Use different dynamic threshold segmentation algorithms to automatically segment the vegetation index image based on the chromaticity coordinates of the RGB image, and select the optimal vegetation index and the optimal segmentation algorithm according to the segmentation accuracy.

4. The method for eliminating the interference information of the crop canopy temperature according to claim 1, wherein After obtaining the multi-spectral image, thermal image, and RGB image of the farmland, the following steps are further included: screening the optimal dynamic threshold segmentation algorithm based on the farmland thermal image, including the following steps: converting the pixel values of the farmland thermal image into Celsius temperature to generate a temperature image; cropping out an image on the temperature image that includes both crop and ridge feature information; automatically segmenting the temperature image using different dynamic threshold segmentation algorithms to generate a binary image of the segmentation result, and screening out the optimal dynamic threshold segmentation algorithm based on the thermal image according to the segmentation accuracy.

5. The method for eliminating crop canopy temperature interference information according to claim 2, characterized in that The vegetation indices calculated based on the reflectance of the multi-spectral image include the Normalized Difference Vegetation Index (NDVI), the Soil Adjusted Vegetation Index (SAVI), and the Normalized Difference Canopy Shadow Index (NDCSI): Among them, R NIR and R Red represent the reflectance in the near-infrared and red bands, respectively; where L represents the soil adjustment factor with a value of 0.5; Among them, R RE represents the reflectance of the red-edge band, and R RE_max and R RE_min represent the maximum and minimum values of the red-edge reflectance of all pixels in the image, respectively.

6. The method for eliminating the interference information of the crop canopy temperature according to claim 3, characterized in that, The vegetation indices based on the chromaticity coordinates of the RGB image include the Excess Green Index (ExG), the Excess Green minus Excess Red Index (ExGR), and the Green-Red Vegetation Index (GRVI): ExG = 2g - r - b ExR = 1.4r - g ExGR = ExG - ExR where r, g, and b are the chromaticity coordinates respectively: where R, G, and B are the actual pixel values on the RGB image.

7. The method for eliminating crop canopy temperature interference information according to any one of claims 1-6, characterized in that The dynamic threshold segmentation algorithms include the Otsu method (OTSU), the Iterative Self-Organizing Data Analysis Technique (ISODATA), the Maximum Entropy method (MAXENTROPY), the Mean method (MEAN), the Minimum Error method (MINERROR), and the Moments method (MOMENTS).

8. The method for eliminating the interference information of the crop canopy temperature according to claim 5, characterized in that, When the crop is winter wheat, the optimal vegetation index based on the multi-spectral image is the Normalized Difference Canopy Shadow Index (NDCSI), and the optimal dynamic threshold segmentation algorithm is the Otsu method (OTSU).

9. The method for eliminating the interference information of the crop canopy temperature according to claim 6, wherein When the crop is winter wheat, the optimal vegetation index based on the RGB image is the Excess Green minus Excess Red Index (ExGR), and the optimal dynamic threshold segmentation algorithm is the Minimum Error method (MINERROR).

10. The method for eliminating crop canopy temperature interference information according to claim 4, characterized in that, When applied to scenarios related to winter wheat, the optimal dynamic threshold segmentation algorithm based on the thermal image is the Mean method (MEAN).

Images