Soil moisture detection method combining deep learning model optimization precision

Abstract

The invention belongs to the technical field of soil moisture detection, and provides a soil moisture detection method combining deep learning model optimization precision, which comprises the following steps: in a plurality of soil moisture detection periods, carrying out moisture detection on a soil sample detection area divided by each soil sample in a vertical direction to obtain a moisture detection report, and performing misjudgment analysis on the soil sample moisture detection amount corresponding to each soil sample detection area in the moisture detection report to obtain a soil sample false detection area, thereby facilitating error analysis on a detection result. By comparing moisture detection data of different detection areas, different depths and different periods, possible error sources can be found out, a soil sample false detection area can be determined, key training can be performed on the soil sample false detection area in a targeted manner in a deep learning model training process, and the detection accuracy of the soil sample false detection area is improved. And the model pays more attention to how to accurately detect the areas which are easy to misjudge, so that the detection performance of the model in the specific areas is optimized, and the overall detection precision is improved.

Description

Technical Field

[0001] This invention belongs to the field of soil moisture detection technology, specifically a soil moisture detection method that combines deep learning models to optimize accuracy. Background Technology

[0002] Soil moisture testing is a crucial task in many fields, including agriculture, ecological research, and environmental monitoring. Accurate soil moisture data plays an irreplaceable role in rational irrigation, crop growth regulation, soil erosion research, and climate change simulation.

[0003] In existing technologies, due to the complexity and variability of the soil environment, soil moisture monitoring data can vary significantly across different monitoring areas, depths, and cycles over multiple monitoring periods. However, traditional equipment typically only outputs the measured values ​​without in-depth analysis to identify potential sources of error. This makes it difficult to determine which areas might have misjudged the data in practical applications, hindering targeted correction and optimization of the test results.

[0004] Moreover, soil uniformity is one of the important factors affecting the accuracy of moisture detection. In actual soil, there are often large differences in the distribution of impurities and soil porosity in the vertical direction. This non-uniformity will interfere with moisture detection. However, most existing detection technologies do not fully consider these characteristics in the vertical direction of the soil. They only detect soil moisture as a whole and cannot accurately identify the moisture detection error caused by soil non-uniformity.

[0005] Therefore, this invention provides a soil moisture detection method that combines deep learning models to optimize accuracy. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is: A soil moisture detection method that combines deep learning models to optimize accuracy includes: In multiple soil moisture testing cycles, the moisture content of each soil sample is measured vertically within the designated testing area, and a moisture test report is obtained. By performing a misjudgment analysis on the soil moisture content corresponding to each soil sample testing area in the moisture testing report, the misjudged soil sample areas are obtained. Soil homogeneity was tested in each soil sample area based on vertical impurity distribution and vertical soil pore dimensions, and the overlap with the false detection areas of the soil sample was compared to assess the type of false detection of soil moisture. If the type of soil sample moisture false detection is soil sample non-uniform signal, then obtain the moisture misjudgment amount and non-uniformity value corresponding to each soil sample false detection area, perform correlation deep learning, obtain non-uniform moisture correction coefficient, and correct the soil sample false detection area in the moisture detection report according to the non-uniform moisture correction coefficient. The product of the moisture misjudgment value corresponding to the soil sample misdetection area and the non-uniform moisture correction coefficient is multiplied and then summed with the moisture misjudgment value corresponding to the soil sample misdetection area to output the corrected moisture detection value.

[0008] As a further aspect of the present invention, the method for obtaining the moisture test report is as follows: Soil samples are divided into grids to obtain multiple soil sample testing areas. These soil sample testing areas are then vertically and equally divided to obtain upper soil sample testing sub-areas, middle soil sample testing sub-areas, and lower soil sample testing sub-areas. The moisture content of the upper, middle and lower soil sample testing sub-regions is obtained separately, and the average value is calculated to output the soil sample moisture content. The soil moisture content of each soil sample in the upper, middle, and lower soil sample testing sub-areas is statistically summarized from smallest to largest to obtain a moisture content test report.

[0009] A further aspect of this invention is as follows: A misjudgment analysis is performed on the soil moisture content measured in each soil sample testing area within the moisture testing report, comparing it with the soil moisture content measured in the adjacent diagonal soil sample testing areas. The process is as follows: Select any soil sample testing area as a target analysis area. Mark other soil sample testing areas adjacent to the target analysis area in the horizontal and vertical dimensions as water-adjacent soil sample testing areas. Obtain the soil moisture content of each adjacent soil sample testing area and the soil moisture content of the target analysis area, calculate the difference, obtain the diagonal water adjacent difference value, calculate the standard deviation, and output the diagonal adjacent water standard deviation.

[0010] A further aspect of this invention is as follows: A misjudgment analysis is performed on the soil moisture content measured in each soil sample testing area within the moisture testing report, comparing it with the moisture content measured in adjacent horizontal soil sample testing areas. The process is as follows: Other soil sample detection areas that are diagonally opposite the target analysis area are marked as corner neighbor soil sample detection areas. The soil moisture content of each water neighbor soil sample detection area is obtained and subtracted from the soil moisture content of the target analysis area to obtain the water neighbor moisture difference value. The standard deviation is then calculated and the water neighbor moisture standard deviation is output.

[0011] As a further aspect of the present invention, the process for identifying false detection areas in soil samples is as follows: The standard deviations of moisture in the corner and the water are summed to obtain the target misjudgment analysis value. If the target misjudgment analysis value is greater than or equal to the target misjudgment analysis threshold, the target analysis area is marked as the soil sample misdetection area.

[0012] A further aspect of this invention is as follows: Soil uniformity is detected in each soil sample testing area from the perspective of vertical impurity distribution, as follows: The volume of impurities in the upper soil sample testing sub-region within the soil sample testing area is obtained, summed to obtain the total volume of impurities, and then the ratio of this volume to the total volume of soil in the upper soil sample testing sub-region is calculated to obtain the volume ratio of impurities in the upper soil sample. The volume of impurities in the soil sample testing sub-region within the soil sample testing area is obtained, summed to obtain the total volume of impurities, and then the ratio of this volume to the total volume of soil in the soil sample testing sub-region is calculated to obtain the volume ratio of impurities in the soil sample. The volume of impurities in the sub-region of the soil sample test area is obtained, summed to obtain the total volume of impurities, and then the ratio of this volume to the total volume of soil in the sub-region of the soil sample test area is calculated to obtain the volume ratio of impurities in the sub-region of the soil sample. The standard deviation of the impurity volume ratios in the upper, middle, and lower soil samples was calculated to obtain the regional impurity mean.

[0013] A further aspect of this invention is as follows: Soil uniformity is measured in each soil sample testing area from the vertical soil pore dimension, as follows: CT scans were performed on the upper, middle, and lower soil sample testing sub-regions within each soil sample testing area to obtain the upper, middle, and lower soil sample vertical profiles. The soil detection vertical profile is divided into several soil vertical profile areas. The volume of each pore in each soil vertical profile area is obtained, and the sum is calculated. The ratio of the summation is then calculated with the total volume of the soil vertical profile area to obtain the pore volume ratio. The vertical profile of the soil test is divided into several vertical profile areas. The volume of each pore in each vertical profile area is obtained, the sum is calculated, and the ratio of the sum to the total volume of the vertical profile area is calculated to obtain the pore volume ratio. The soil detection vertical profile is divided into several soil vertical profile areas. The volume of each pore in each soil vertical profile area is obtained, the volumes are summed, and the ratio of the summation to the total volume of the soil vertical profile area is calculated to obtain the pore volume ratio. The standard deviation of the volume ratios of the upper, middle, and lower pores is calculated to obtain the mean porosity of the region.

[0014] A further aspect of this invention is as follows: identifying non-uniform detection areas and comparing them with false detection areas in the soil sample, the process of which is as follows: The average value of impurities in the region is summed with the average value of porosity in the region to obtain the uniform detection value of the region. If the uniform detection value of the region is greater than the uniform detection threshold of the region, the soil sample detection area is marked as a non-uniform detection area. The spatial location of the non-uniform detection area is compared with that of the soil sample false detection area. The number of overlapping areas is counted and the ratio of the number of soil sample detection areas is calculated to obtain the comparison overlap value. If the comparison overlap value is greater than or equal to the comparison overlap threshold, it indicates that the spatial overlap between the non-uniform detection area and the soil sample false detection area is high, which is displayed as a soil non-uniformity signal.

[0015] A further aspect of this invention is as follows: The amount of moisture misjudgment and the degree of heterogeneity corresponding to each soil sample's misjudgment area are obtained, as follows: The mean of the differences between all diagonal water levels is calculated, and the mean of the misjudged water level is output. The mean of the differences between all adjacent water levels is calculated, and the mean of the misjudged water level is output. The mean of the misjudged water level is summed with the mean of the misjudged water level, and the misjudged water level is output. The difference between the uniform detection value of the region and the uniform detection threshold of the region is used to obtain the degree of non-uniformity.

[0016] A further aspect of this invention is as follows: the process of obtaining the uneven moisture correction coefficient through association-based deep learning is as follows: The mean and standard deviation of the moisture misjudgment amounts within the moisture misjudgment sequence are calculated, and the mean and standard deviation of the moisture misjudgment amounts are output. The mean and standard deviation of the non-uniformity values ​​within the non-uniformity sequence are calculated separately, and the mean and standard deviation of the non-uniformity are output. The mean of moisture misjudgment, the standard deviation of moisture misjudgment, the mean of non-uniformity, and the standard deviation of non-uniformity are calculated using the Pearson correlation coefficient formula, and the absolute value is taken to output the correlation depth coefficient. If the correlation depth coefficient approaches 1, then the ratio of the non-uniformity value to the moisture misjudgment value within each correlation depth learning group is calculated, and the mean is calculated to output the non-uniform moisture correction coefficient.

[0017] The beneficial effects of this invention are as follows: (1) In multiple soil moisture detection cycles, the present invention performs vertical moisture detection on the soil sample detection area divided by each soil sample to obtain a moisture detection report. The soil moisture detection amount corresponding to each soil sample detection area in the moisture detection report is analyzed for misjudgment to obtain the soil sample misdetection area, which helps to analyze the error of the detection results. By comparing the moisture detection data of different detection areas, different depths and different cycles, the possible sources of error can be found and the soil sample misdetection area can be determined. In the deep learning model training process, the soil sample misdetection area can be trained in a targeted manner, so that the model pays more attention to how to accurately detect these areas that are easy to misjudge, thereby optimizing the detection performance of the model in these specific areas and improving the overall detection accuracy. (2) This invention detects soil uniformity in each soil sample detection area from the perspectives of vertical impurity distribution and vertical soil pores, and compares it with the soil sample false detection area to evaluate the type of soil sample moisture false detection. If the soil sample moisture false detection type is soil sample non-uniformity signal, the moisture misjudgment amount and non-uniformity value corresponding to each soil sample false detection area are obtained, and correlation deep learning is performed to obtain the non-uniform moisture correction coefficient. Thus, for soil samples marked as non-uniform detection areas with large moisture misjudgment amounts, their moisture detection values ​​are adjusted according to the non-uniform moisture correction coefficient. This is beneficial for increasing the training intensity in areas with high non-uniformity and large moisture misjudgment amounts, optimizing the detection performance of the model in these specific areas, and improving the overall detection accuracy. Moreover, by considering the influence of soil non-uniformity on moisture detection and correcting the detection data, the model can resist these interference factors to a certain extent, maintain the stability of the detection results, and help adapt to various complex soil environments, thereby improving the model's generalization ability. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart of the steps of a soil moisture detection method that combines deep learning models to optimize accuracy according to the present invention. Figure 2 This is a flowchart illustrating the judgment process of a soil moisture detection method that combines deep learning models to optimize accuracy, according to the present invention. Figure 3 This is a schematic diagram of a module within a soil moisture detection system that combines a deep learning model to optimize accuracy, according to the present invention. Detailed Implementation

[0020] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0021] Example 1 Please see Figures 1-2As shown in the embodiment of the present invention, a soil moisture detection method that combines deep learning models to optimize accuracy includes the following methods: Step 1: During multiple soil moisture testing cycles, perform vertical moisture testing on the soil sample testing area of ​​each soil sample to obtain a moisture testing report. It should be noted that the soil moisture testing cycle is the period during which soil samples are tested using soil moisture testing equipment, and only one soil sample is tested within the soil moisture testing cycle. In some embodiments, the soil sample is divided into grids to obtain multiple soil sample detection areas; It should be noted that each soil sample testing area is equal in both vertical depth and horizontal area. The process of vertical moisture detection for soil sample testing areas is as follows: For example: The soil sample testing area is vertically and equally divided to obtain soil sample testing sub-areas, wherein the soil sample testing sub-areas include an upper soil sample testing sub-area, a middle soil sample testing sub-area, and a lower soil sample testing sub-area; Among them, the vertical heights of the upper soil sample testing sub-area, the middle soil sample testing sub-area, and the lower soil sample testing sub-area are all equal; The moisture content of the upper, middle and lower soil sample testing sub-regions is obtained separately, and the average value is calculated to output the soil sample moisture content. The soil moisture content of each upper, middle and lower soil sample testing sub-area is statistically summarized from smallest to largest to obtain a moisture test report. Understandably, the purpose of obtaining a moisture test report is: Objective 1: The moisture test report records the vertical moisture content information of each soil sample test area, which can serve as an important basis for subsequent analysis and optimization of the deep learning model. Moreover, the deep learning model itself needs a large amount of real data to learn the pattern of soil moisture distribution and possible error factors. The data in the moisture test report can provide the model with real and specific samples. Objective 2: From the perspective of detection accuracy, the detailed data records in the moisture detection report help to analyze the error of the detection results. By comparing moisture detection data from different detection areas, different depths, and different periods, it is possible to identify possible sources of error, such as the accuracy error of the detection equipment, errors caused by soil heterogeneity, etc. Moreover, deep learning models can be used to learn how to identify and correct errors using these error analysis results, thereby improving the accuracy of soil moisture detection.

[0022] Step 2: Perform a misjudgment analysis on the soil moisture content corresponding to each soil sample testing area in the moisture test report to obtain the misjudged soil sample areas; In some embodiments, the horizontal analysis of the soil sample testing area is performed as follows: Select any soil sample testing area as a target analysis area, and mark other soil sample testing areas that are adjacent to the target analysis area in the horizontal and vertical dimensions as water-adjacent soil sample testing areas. Other soil sample detection areas that are diagonally opposite the target analysis area in space are marked as corner neighbor soil sample detection areas; For example, for the corner neighbor soil sample detection area, the soil moisture detection amount corresponding to the target analysis area is subtracted from the soil moisture detection amount corresponding to each corner neighbor soil sample detection area to obtain the diagonal moisture adjacent difference value, and the standard deviation is calculated to output the corner neighbor moisture standard deviation. Similarly, for the soil sample testing area adjacent to the water, the difference between the soil sample moisture detection amount corresponding to the target analysis area and the soil sample moisture detection amount corresponding to each soil sample testing area adjacent to the water is obtained, and the standard deviation is calculated and the standard deviation of the water sample moisture is output. The standard deviations of moisture in the corner neighbors and the standard deviations of moisture in the water neighbors are summed to output the target misjudgment analysis value; It is understandable that the target misjudgment analysis value represents a comprehensive value obtained by comparing the moisture content of the target analysis area with that of adjacent soil sample detection areas and adjacent water sample detection areas. Specifically, the standard deviation of moisture content in the corner area reflects the dispersion of moisture content between the target analysis area and its diagonally adjacent areas; by reflecting the dispersion of moisture content between the target analysis area and its horizontal and vertical adjacent areas, it comprehensively measures the degree of difference in moisture content between the target analysis area and its surrounding adjacent areas in multiple horizontal and spatial diagonal directions. The target misclassification analysis value is compared with the target misclassification analysis threshold, as follows: If the target misjudgment analysis value is greater than or equal to the target misjudgment analysis threshold, it indicates that the moisture detection amount in the target analysis area is significantly different from that in the surrounding adjacent areas, and the target analysis area is marked as a soil sample misjudgment area. If the target misjudgment analysis value is less than the target misjudgment analysis threshold, it indicates that the difference in moisture detection between the target analysis area and the surrounding adjacent areas is small, and the target analysis area is marked as the soil sample positive detection area. It should be noted that the target misjudgment analysis threshold refers to the following: Since soil samples are divided into multiple soil sample testing areas using a grid method, and each testing area is equal in the horizontal area dimension, if there is an unreasonable and large difference in the detection amount of soil sample testing areas at different horizontal positions, there may be a misjudgment. For example, at the same horizontal level, the detection amounts of several surrounding testing areas are similar, but the detection amount of a certain testing area is abnormally high or low. Therefore, the target misjudgment analysis threshold is obtained by subtracting the detection amounts of adjacent areas from each other and averaging the differences. The significance of using target misclassification analysis values ​​to screen for false detection areas in multiple soil sample testing areas is that: From the perspective of learning model training, after identifying the false detection areas of soil samples, the deep learning model training process can focus on training the false detection areas of soil samples, making the model pay more attention to how to accurately detect these areas that are prone to misjudgment, thereby optimizing the detection performance of the model in these specific areas and improving the overall detection accuracy. In terms of improving detection accuracy, identifying and specifically addressing areas of soil sample misdetection can significantly enhance the accuracy of soil moisture detection in these areas. Furthermore, when detecting the moisture content of each soil sample area, the overall distribution of soil moisture can be more accurately assessed. This is of great significance for studying the dynamic changes in soil moisture and developing appropriate irrigation strategies, providing more reliable data support for agricultural production and other related fields.

[0023] The specific solution in this embodiment is as follows: During multiple soil moisture testing cycles, vertical moisture testing is performed on the soil sample testing areas divided for each soil sample to obtain a moisture testing report. A misjudgment analysis is then conducted on the soil moisture measured in each soil sample testing area within the moisture testing report to identify the misjudged areas. This helps in error analysis of the testing results. By comparing moisture testing data from different testing areas, depths, and cycles, potential sources of error can be identified, and misjudged areas can be determined. During the deep learning model training process, targeted training can be conducted on these misjudged areas, making the model more focused on accurately detecting these easily misjudged areas, thereby optimizing the model's detection performance in these specific areas and improving overall detection accuracy.

[0024] Example 2 Please see Figures 1-2 As shown, in addition to Embodiment 1, this embodiment of the invention also includes the following method: Step 3: Detect soil homogeneity in each soil sample testing area based on vertical impurity distribution and vertical soil pore dimensions, and compare it with the false detection areas of the soil sample to assess the type of false detection of soil moisture. Among them, the types of false detection of soil moisture include equipment misjudgment signals or soil heterogeneity signals; In some embodiments, soil homogeneity is tested in each soil sample testing area from the perspective of vertical impurity distribution, as follows: For example, the volume of impurities in the upper soil sample testing sub-region within the soil sample testing area is obtained, summed to obtain the total volume of impurities, and the ratio of this volume to the total volume of soil in the upper soil sample testing sub-region is calculated to obtain the volume ratio of impurities in the upper soil sample. The volume of impurities in the soil sample testing sub-region within the soil sample testing area is obtained, summed to obtain the total volume of impurities, and then the ratio of this volume to the total volume of soil in the soil sample testing sub-region is calculated to obtain the volume ratio of impurities in the soil sample. The volume of impurities in the sub-region of the soil sample test area is obtained, summed to obtain the total volume of impurities, and then the ratio of this volume to the total volume of soil in the sub-region of the soil sample test area is calculated to obtain the volume ratio of impurities in the sub-region of the soil sample. The standard deviation of the volume ratio of impurities in the upper soil sample, the middle soil sample, and the lower soil sample is calculated to obtain the regional impurity mean. Soil homogeneity was assessed for each soil sample area along the vertical soil pore dimension, as follows: CT scans were performed on the upper, middle, and lower soil sample testing sub-regions within each soil sample testing area to obtain the upper, middle, and lower soil sample vertical profiles. For example, the soil detection vertical profile is equally divided into several soil vertical profile areas, the volume of each pore in each soil vertical profile area is obtained, the sum is calculated, and the ratio is calculated with the total volume of the soil vertical profile area to obtain the pore volume ratio. The vertical profile of the soil test is divided into several vertical profile areas. The volume of each pore in each vertical profile area is obtained, the sum is calculated, and the ratio of the sum to the total volume of the vertical profile area is calculated to obtain the pore volume ratio. The soil detection vertical profile is divided into several soil vertical profile areas. The volume of each pore in each soil vertical profile area is obtained, the volumes are summed, and the ratio of the summation to the total volume of the soil vertical profile area is calculated to obtain the pore volume ratio. The standard deviation of the volume ratio of the upper pore, the volume ratio of the middle pore, and the volume ratio of the lower pore is calculated to obtain the mean porosity of the region. The average value of regional impurities is summed with the average value of regional porosity to obtain the uniform detection value of the region. It is understandable that the meaning of the regional uniformity test value is: to reflect the uniformity of the soil sample test area in the vertical direction in terms of two key dimensions: impurity distribution and soil porosity. On the one hand, the regional impurity mean reflects the dispersion of impurity distribution in the vertical direction of the soil sample test area; on the other hand, the regional porosity mean reflects the dispersion of soil porosity distribution in the vertical direction of the soil sample test area. The region-uniform detection value is compared with the region-uniform detection threshold, as follows: If the regional uniform detection value is greater than the regional uniform detection threshold, it indicates that the impurities and soil porosity in the upper, middle and lower sub-regions are significantly different, and the soil sample detection area is marked as a non-uniform detection area. If the regional uniform detection value is less than or equal to the regional uniform detection threshold, it indicates that the differences in impurities and soil porosity among the upper, middle and lower sub-regions are small, and the soil sample detection area is marked as a uniform detection area. It should be noted that the regional uniformity detection threshold refers to: The spatial location of the non-uniform detection area is compared with that of the soil sample false detection area. The number of overlapping areas is counted and the ratio of the number of soil sample detection areas is calculated to obtain the comparison overlap value. If the comparison overlap value is greater than or equal to the comparison overlap threshold, it indicates that the spatial overlap between the non-uniform detection area and the soil sample false detection area is high, which is a soil non-uniform signal. If the comparison overlap value is less than the comparison overlap threshold, it indicates that the spatial overlap between the non-uniform detection area and the soil sample misdetection area is low, which is a signal of equipment misjudgment. The significance of assessing the false detection types of soil moisture is that by assessing the false detection types of soil moisture, we can accurately distinguish whether the error is caused by the problem of the detection equipment itself or by the heterogeneity of the soil itself. By collecting more soil sample data under different soil types, different impurities and pore distributions, the model can better adapt to the detection challenges brought about by soil heterogeneity, thereby improving the model's ability to identify and process different error situations. Different types of false detection signals reflect different problem essences. The model can learn the deviation patterns that the equipment may exhibit under different environments and usage conditions, thereby automatically correcting equipment errors during the detection process and improving the accuracy of detection results. Step 4: If the type of soil sample moisture false detection is soil sample non-uniform signal, then obtain the moisture misjudgment amount and non-uniformity value corresponding to each soil sample false detection area, perform correlation deep learning, obtain non-uniform moisture correction coefficient, and correct the soil sample false detection area in the moisture detection report according to the non-uniform moisture correction coefficient. In some embodiments, the average difference between adjacent diagonal moisture values ​​is calculated to output the average value of misjudged diagonal moisture. Calculate the average of all adjacent water level differences and output the average of the water level misclassifications. The mean of misjudged water content at the corner is summed with the mean of misjudged water content at the adjacent water level, and the misjudged water content is output as the amount of misjudged water content. The difference between the uniform detection value of the region and the uniform detection threshold of the region is used to obtain the degree of non-uniformity. The misjudgment of moisture and the degree of non-uniformity corresponding to each soil sample are combined to obtain multiple sets of correlation in-depth learning groups; The number of water misjudgments within each group of association-deep learning groups is compared and sorted in descending order to obtain the water misjudgment sequence; Similarly, the non-uniformity values ​​within each group of association study groups are compared and sorted in descending order to obtain a non-uniformity sequence. The mean and standard deviation of the moisture misjudgment amounts within the moisture misjudgment sequence are calculated, and the mean and standard deviation of the moisture misjudgment amounts are output. The mean and standard deviation of the non-uniformity values ​​within the non-uniformity sequence are calculated separately, and the mean and standard deviation of the non-uniformity are output. The mean of moisture misjudgment, the standard deviation of moisture misjudgment, the mean of non-uniformity, and the standard deviation of non-uniformity are calculated using the Pearson correlation coefficient formula, and the absolute value is taken to output the correlation depth coefficient. If the correlation coefficient approaches 0, no action is taken. If the correlation depth coefficient approaches 1, then the ratio of the non-uniformity value to the moisture misjudgment value in each correlation depth learning group is calculated, and the mean is calculated to output the non-uniform moisture correction coefficient. The product of the misjudged moisture value corresponding to the false detection area of ​​the soil sample and the non-uniform moisture correction coefficient is multiplied and then summed with the misjudged moisture value corresponding to the false detection area of ​​the soil sample to output the corrected moisture detection value. The significance of obtaining the non-uniform moisture correction coefficient is as follows: From the perspective of data correction, since different soil samples have different impurity distributions and soil porosity, the corresponding non-uniform moisture correction coefficients are also different. Therefore, for soil samples marked as non-uniform detection areas with large moisture misjudgments, adjusting their moisture detection values ​​according to the non-uniform moisture correction coefficient makes the data closer to the true value. This is beneficial for increasing the training intensity for areas with high non-uniformity and large moisture misjudgments, making the model more focused on how to accurately detect these areas that are prone to misjudgment due to soil heterogeneity, thereby optimizing the model's detection performance in these specific areas and improving the overall detection accuracy. From the perspective of model adaptability, in actual soil moisture detection environments, various factors may interfere, such as the accuracy error of detection equipment and changes in environmental conditions. The heterogeneous moisture correction coefficient provides a means for the model to cope with these interferences. By considering the impact of soil heterogeneity on moisture detection and correcting the detection data, the model can resist these interference factors to a certain extent, maintain the stability of the detection results, and help adapt to various complex soil environments, thereby improving the model's generalization ability. The specific scheme of this embodiment is as follows: Soil homogeneity is detected in each soil sample detection area from the dimensions of vertical impurity distribution and vertical soil pores, and the overlap with the soil sample false detection area is compared to evaluate the type of soil sample moisture false detection. If the soil sample moisture false detection type is soil inhomogeneity signal, the moisture misjudgment amount and non-homogeneity value corresponding to each soil sample false detection area are obtained, and correlation deep learning is performed to obtain the non-homogeneity moisture correction coefficient. Thus, for soil samples marked as non-homogeneity detection areas with large moisture misjudgment amounts, their moisture detection values ​​are adjusted according to the non-homogeneity moisture correction coefficient. This is beneficial for increasing the training intensity in areas with high non-homogeneity and large moisture misjudgment amounts, optimizing the model's detection performance in these specific areas, improving the overall detection accuracy, and by considering the influence of soil inhomogeneity on moisture detection and correcting the detection data, the model can resist these interference factors to a certain extent, maintain the stability of the detection results, and help adapt to various complex soil environments, improving the model's generalization ability.

[0025] Example 3

[0026] Based on the same inventive concept as the soil moisture detection method that combines a deep learning model to optimize accuracy in the aforementioned embodiments, such as Figure 3 As shown, this application provides a soil moisture detection system that combines a deep learning model to optimize accuracy. Specifically, the system includes the following modules: Soil Moisture Testing Module: During multiple soil moisture testing cycles, the moisture content of each soil sample is measured vertically within the designated testing area, and a moisture testing report is generated. By performing a misjudgment analysis on the soil moisture content corresponding to each soil sample testing area in the moisture testing report, the misjudged soil sample areas are obtained. False detection area identification module: Detects soil uniformity in each soil sample detection area based on vertical impurity distribution and vertical soil pore dimension, and compares it with the false detection area of ​​the soil sample to assess the type of soil sample moisture false detection. False detection type analysis module: If the false detection type of soil sample moisture is soil sample non-uniform signal, then obtain the moisture misjudgment amount and non-uniformity value corresponding to each soil sample false detection area, perform correlation in-depth learning, obtain non-uniform moisture correction coefficient, and correct the soil sample false detection area in the moisture detection report according to the non-uniform moisture correction coefficient. The precision correction module calculates the corrected moisture detection value by multiplying the moisture misjudgment value corresponding to the soil sample misdetection area with the non-uniform moisture correction coefficient, and then summing the product with the moisture misjudgment value corresponding to the soil sample misdetection area.

[0027] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A soil moisture detection method combining a deep learning model to optimize precision, characterized in that: The method comprises the following steps: During multiple soil moisture detection periods, vertical moisture detection is performed on each soil sample detection area of each soil sample to obtain a moisture detection report; False detection analysis is performed on the soil moisture detection amount corresponding to each soil sample detection area in the moisture detection report to obtain a soil sample false detection area; Soil uniformity detection is performed on each soil sample detection area in the vertical impurity distribution and vertical soil pore dimension, and is compared with the soil sample false detection area to evaluate the type of soil moisture false detection; If the type of soil moisture false detection is soil non-uniformity signal, the false detection amount of each soil sample false detection area and the non-uniformity degree value are obtained for correlation learning to obtain a non-uniformity moisture correction coefficient, and the soil sample false detection area in the moisture detection report is corrected according to the non-uniformity moisture correction coefficient; The product of the false detection amount of the soil sample false detection area and the non-uniformity moisture correction coefficient is summed with the false detection amount of the soil sample false detection area to output the corrected moisture detection amount.

2. The soil moisture detection method of claim 1, wherein: The moisture detection report is obtained by: dividing the soil sample into a grid to obtain multiple soil sample detection areas, and vertically and equally dividing the soil sample detection areas to obtain upper, middle and lower soil sample detection sub-areas; obtaining the moisture content of the upper, middle and lower soil sample detection sub-areas, respectively, and performing mean value calculation to output the soil moisture detection amount; statistically summarizing the soil moisture detection amount corresponding to each upper, middle and lower soil sample detection sub-area from small to large to obtain the moisture detection report.

3. The soil moisture detection method of claim 1, wherein: The difference false detection analysis is performed on the soil moisture detection amount corresponding to each soil sample detection area in the moisture detection report and the soil moisture detection amount corresponding to the adjacent diagonal soil sample detection area as follows: An arbitrary soil sample detection area is selected as a target analysis area, and other soil sample detection areas adjacent to the target analysis area in the spatial horizontal and vertical dimensions are marked as water adjacent soil sample detection areas. The soil moisture detection amount corresponding to each corner adjacent soil sample detection area and the soil moisture detection amount corresponding to the target analysis area are subtracted to obtain the diagonal moisture adjacent difference value, and standard deviation calculation is performed to output the corner adjacent moisture standard deviation.

4. The soil moisture detection method of claim 1, wherein: The difference false detection analysis is performed on the soil moisture detection amount corresponding to each soil sample detection area in the moisture detection report and the soil moisture detection amount corresponding to the adjacent horizontal soil sample detection area as follows: The other soil sample detection areas adjacent to the target analysis area in the spatial diagonal dimension are marked as corner adjacent soil sample detection areas. The soil moisture detection amount corresponding to each water adjacent soil sample detection area and the soil moisture detection amount corresponding to the target analysis area are subtracted to obtain the water adjacent moisture adjacent difference value, and standard deviation calculation is performed to output the water adjacent moisture standard deviation. 5.The soil moisture detection method of claim 1, wherein: The identification process of the soil sample false detection area is as follows: The corner adjacent moisture standard deviation and the water adjacent moisture standard deviation are summed to output a target false detection analysis value. If the target false detection analysis value is greater than or equal to a target false detection analysis threshold, the target analysis area is marked as a soil sample false detection area. 6.The soil moisture detection method of claim 1, wherein: From the vertical impurity distribution dimension, the soil homogeneity of each soil sample detection area is detected, and the process is as follows: The volume of impurities in the upper soil sample detection sub-area in the soil sample detection area is obtained, summed up to obtain the total volume of impurities, and the total volume of soil in the upper soil sample detection sub-area is calculated by ratio to obtain the volume ratio of impurities in the upper soil sample; The volume of impurities in the middle soil sample detection sub-area in the soil sample detection area is obtained, summed up to obtain the total volume of impurities, and the total volume of soil in the middle soil sample detection sub-area is calculated by ratio to obtain the volume ratio of impurities in the middle soil sample; The volume of impurities in the lower soil sample detection sub-area in the soil sample detection area is obtained, summed up to obtain the total volume of impurities, and the total volume of soil in the lower soil sample detection sub-area is calculated by ratio to obtain the volume ratio of impurities in the lower soil sample; The volume ratio of impurities in the upper soil sample, the volume ratio of impurities in the middle soil sample and the volume ratio of impurities in the lower soil sample are calculated by standard deviation to obtain the average value of impurities in the region.

7. The soil moisture detection method of claim 1, wherein: From the vertical soil pore dimension, the soil homogeneity of each soil sample detection area is detected, and the process is as follows: CT scanning is performed on the upper soil sample detection sub-area, the middle soil sample detection sub-area and the lower soil sample detection sub-area in each soil sample detection area respectively to obtain the upper soil detection vertical profile, the middle soil detection vertical profile and the lower soil detection vertical profile; The upper soil detection vertical profile is equally divided into a plurality of upper soil vertical profile areas, the volume of each pore in each upper soil vertical profile area is obtained, summed up, and the total volume of the upper soil vertical profile area is calculated by ratio to obtain the upper pore volume ratio; The middle soil detection vertical profile is equally divided into a plurality of middle soil vertical profile areas, the volume of each pore in each middle soil vertical profile area is obtained, summed up, and the total volume of the middle soil vertical profile area is calculated by ratio to obtain the middle pore volume ratio; The lower soil detection vertical profile is equally divided into a plurality of lower soil vertical profile areas, the volume of each pore in each lower soil vertical profile area is obtained, summed up, and the total volume of the lower soil vertical profile area is calculated by ratio to obtain the lower pore volume ratio; The upper pore volume ratio, the middle pore volume ratio and the lower pore volume ratio are calculated by standard deviation to obtain the average value of pores in the region. 8.The soil moisture detection method of claim 5, wherein: The non-homogeneous detection area is identified and compared with the soil sample false detection area, and the process is as follows: The sum of the average value of impurities in the region and the average value of pores in the region is obtained to obtain the uniform detection value of the region, and if the uniform detection value of the region is greater than the uniform detection threshold value of the region, the soil sample detection area is marked as a non-homogeneous detection area; The non-homogeneous detection area and the soil sample false detection area are compared in space position, the number of compared overlapping areas is counted, and the ratio of the total number of soil sample detection areas is calculated to output the compared overlapping value; If the compared overlapping value is greater than or equal to the compared overlapping threshold value, it indicates that the spatial coincidence degree of the non-homogeneous detection area and the soil sample false detection area is high, and the soil non-homogeneous signal is displayed. 9.The soil moisture detection method of claim 1, wherein: The moisture misjudgment amount corresponding to each soil sample false detection area and the non-homogeneous degree value are obtained, and the process is as follows: The average value of all diagonal water adjacent difference values is calculated to output the angle water misjudgment average value, the average value of all water adjacent water adjacent difference values is calculated to output the water adjacent water misjudgment average value, and the sum of the angle water misjudgment average value and the water adjacent water misjudgment average value is calculated to output the water misjudgment amount; The non-uniformity degree value is obtained by subtracting the area uniformity detection threshold from the area uniformity detection value.

10. The soil moisture detection method of claim 9, wherein: The process of obtaining the non-uniformity moisture correction coefficient through correlation deep learning is as follows: The moisture misjudgment quantity in the moisture misjudgment sequence is subjected to mean value and standard deviation calculation respectively, and the moisture misjudgment mean value and the moisture misjudgment standard deviation are output. The non-uniformity degree value in the non-uniformity degree sequence is subjected to mean value and standard deviation calculation respectively, and the non-uniformity degree mean value and the non-uniformity degree standard deviation are output. The moisture misjudgment mean value, the moisture misjudgment standard deviation, the non-uniformity degree mean value and the non-uniformity degree standard deviation are calculated through the Pearson correlation coefficient formula, and the absolute value is taken, and the correlation deep learning coefficient is output. If the correlation deep learning coefficient tends to 1, the non-uniformity degree value and the moisture misjudgment quantity in each correlation deep learning learning group are subjected to ratio calculation, and then the mean value is calculated, and the non-uniformity moisture correction coefficient is output.

Images