Soil fertility intelligent evaluation method and system based on image recognition

Abstract

The invention discloses a soil fertility intelligent assessment method and system based on image recognition, and relates to the technical field of data processing, and the method comprises the steps: obtaining land image information of a to-be-assessed region, the land image information comprising a multispectral image, a thermal imaging image and an RGB image of the to-be-assessed region; carrying out image preprocessing on the land image information, and carrying out vegetation coverage and health assessment according to the multispectral image: speculating soil characteristics according to the RGB image: speculating the humidity change of a soil surface layer according to the temperature distribution of a thermal imaging image, and identifying a drought region and a region with sufficient moisture; integrating the high-resolution texture features of the RGB image, the spectral information of the multispectral image and the speculation result of the temperature distribution of the thermal imaging image to generate a comprehensive analysis result; and classifying soil types and evaluating fertility grades according to a comprehensive analysis result. The problem that in the prior art, in intelligent soil fertility evaluation, workers need to carry out on-site sampling examination is solved, and the technical effect of improving the soil fertility evaluation efficiency is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a soil fertility intelligent evaluation method and system based on image recognition. Background Art

[0002] Soil fertility refers to the ability of soil to continuously provide and regulate the air, temperature, nutrients and non-toxic substances required for plant growth. Soil fertility is affected by many factors, such as soil texture, structure, moisture conditions, temperature, biological conditions, organic matter content and pH value. In order to ensure the healthy growth of plants, growers need to monitor and adjust the soil fertility to achieve an appropriate level.

[0003] At present, most existing technologies rely on mathematical algorithms or mathematical models to analyze multiple evaluation indicators to determine soil fertility or soil quality. These methods usually collect soil samples and conduct experimental analysis, and then calculate soil fertility based on the acquired data. However, these methods have some limitations. First, they usually do not involve how to track long-term changes in soil fertility, but only analyze soil samples at a certain point in time. If the method of sampling and analyzing or fitting at multiple time points is adopted, it will not only be time-consuming and labor-intensive, but also increase costs.

[0004] Another common solution is to use the common sampling method, that is, to collect soil samples at a certain part of the plot and use the analysis results as a representative of the soil fertility of the entire plot. However, this method can usually only reflect the soil conditions in a local area, which can easily lead to one-sided sample data and affect the accuracy of soil fertility assessment. In addition, some technologies also use neural network algorithms and machine learning methods to predict soil fertility. These methods rely on a large amount of sample data for model training. However, when the sample data is insufficient, the prediction accuracy of the model is often low. At the same time, for some areas where the plots are far apart and sampling is difficult, the working efficiency of the existing methods is also low. Summary of the invention

[0005] In order to overcome the defects of the prior art in terms of sample limitation, time cost and data dependence, the present application provides a soil fertility intelligent assessment method and system based on image recognition, which not only overcomes the above defects but also improves the efficiency and accuracy of soil fertility assessment through intelligent means. The specific scheme is as follows: The first aspect of the present application provides a soil fertility intelligent assessment method and system based on image recognition, the method comprising: Identify the area to be assessed; Obtaining land image information of the area to be evaluated, wherein the land image information includes a multispectral image, a thermal imaging image, and an RGB image of the area to be evaluated; Performing image preprocessing on the land image information to remove various interference factors; Performing vegetation coverage and health assessment based on the multispectral image: calculating the vegetation index of the area to be assessed through the multispectral image, and inferring the vegetation health status and soil organic matter content of the area to be assessed based on the vegetation index of the area; Inferring soil characteristics based on the RGB image: extracting texture features of the soil surface through the RGB image, analyzing the arrangement, cracks and pore structure of soil particles, so as to infer the stability and porosity of the soil structure; Based on the temperature distribution of thermal imaging images, the moisture changes of the soil surface can be inferred to identify drought areas and areas with sufficient moisture; The high-resolution texture features of RGB images, the spectral information of multispectral images, and the inferred results of temperature distribution of thermal imaging images are integrated to generate comprehensive analysis results; then, soil types are classified and fertility levels are evaluated based on the comprehensive analysis results.

[0006] Furthermore, the land image information is subjected to image preprocessing to remove various interference factors, including: Noise removal: Use Gaussian filtering or median filtering to remove noise from the image to ensure that the extracted features are not interfered by noise.

[0007] Enhance contrast: Use histogram equalization or adaptive histogram equalization to increase the contrast of the image and make the texture features more obvious.

[0008] Image registration: spatially align images from different sources to ensure the accuracy of subsequent feature fusion.

[0009] Furthermore, the vegetation coverage and health assessment is performed based on the multispectral image: the vegetation index of the area to be assessed is calculated through the multispectral image, and the vegetation health status and soil organic matter content of the area to be assessed are inferred based on the vegetation index of the area, including: According to the red and near-infrared bands in the multispectral image, the NDVI value of each pixel is calculated to generate an NDVI image; According to the red, near-infrared and blue light bands, the EVI value of each pixel is calculated to generate an EVI image; Based on NDVI, assess the vegetation coverage of the soil area; The EVI can be used to further differentiate the health of high-density vegetation areas and identify whether vegetation growth is affected by water or nutrient deficiency; Based on the vegetation coverage and health status of the soil area assessed by the levels of NDVI and EVI, the vegetation health status and soil organic matter content of the area to be assessed are inferred.

[0010] Further, the soil characteristics are inferred based on the RGB image: the texture characteristics of the soil surface are extracted through the RGB image, and the arrangement, cracks and pore structure of the soil particles are analyzed to infer the stability and porosity of the soil structure, including: Extract texture features based on gray-level co-occurrence matrix (GLCM) from RGB images; The local texture pattern of the image is extracted using the LBP method, and the LBP features are represented in the form of a histogram; Use wavelet transform to perform frequency decomposition on the image and extract texture features at different scales; By combining the analysis results of GLCM, LBP and wavelet transform, the soil structure inference results are obtained.

[0011] Furthermore, the high-resolution texture features of the RGB image, the spectral information of the multispectral image, and the inferred results of the temperature distribution of the thermal imaging image are integrated to generate comprehensive analysis results, including: Perform weighted voting on the independent analysis results of RGB images, multispectral images, and thermal images, with each result being assigned a different weight based on its credibility or importance; Set decision rules and integrate the results of each independent analysis according to the predetermined rules; Use ensemble learning methods such as decision trees or random forests to take analysis results from different sources as input features and generate a comprehensive analysis result.

[0012] Furthermore, the soil type classification and fertility level assessment based on the comprehensive analysis results include: Use soil sample images with known fertility levels to annotate the model and generate a training dataset; Use machine learning or deep learning models to perform classification or regression analysis on the comprehensive analysis results to predict soil fertility.

[0013] The second aspect of the present application provides a soil fertility intelligent assessment system based on image recognition, the system comprising: An intelligent soil fertility assessment system based on image recognition, the system comprising: Image acquisition module: The image acquisition module is responsible for acquiring land images of the area to be evaluated, including RGB images, multispectral images and thermal imaging images, and transmitting the acquired image data to the image preprocessing module for processing; Image preprocessing module: preprocesses the collected images, including denoising, correction, image registration and resolution adjustment, and transmits the processed image data to the feature extraction and analysis module as input for feature analysis; Feature extraction and analysis module: extracts key features from different types of images and analyzes them, and transmits the analysis results to the multi-source information fusion module; RGB images: extract texture features and analyze soil structure stability and porosity; Multispectral images: calculate vegetation index, assess vegetation coverage and health, and infer soil organic matter content and nutrient status; Thermal imaging: Analyze temperature distribution, infer soil moisture, and identify drought and well-watered areas; Multi-source information fusion module: fuses feature data from different image sources to form a comprehensive soil feature description; Specifically, weighted voting, rule integration or integrated learning model is used to combine the independent analysis results of RGB images, multispectral images and thermal imaging images to generate comprehensive analysis results, and the fused comprehensive analysis results are transmitted to the soil fertility assessment module; Soil Fertility Assessment Module: Based on the integrated analysis results, soil type classification and fertility level assessment are carried out.

[0014] Specifically, through machine learning or deep learning: use the trained model to perform classification or regression analysis on the comprehensive analysis results to predict the soil fertility level.

[0015] One or more technical solutions provided in this application have at least the following technical effects or advantages: 1. Global coverage: Through the analysis of RGB images, multispectral images and thermal imaging images, the soil conditions of the entire plot can be covered, no longer limited to local samples, greatly improving the comprehensiveness and accuracy of the assessment.

[0016] 2. Efficient analysis: Using image processing and machine learning technology, the dynamic changes of soil can be analyzed in real time, reducing time costs and improving work efficiency.

[0017] 3. Data fusion: The system uses multi-source information fusion technology to integrate the features of different image data and generate comprehensive analysis results, avoiding dependence on a single data source and ensuring the reliability of the evaluation results.

[0018] 4. Intelligent prediction: It integrates intelligent algorithms and can accurately predict the development trend of soil fertility through comprehensive analysis of historical data and real-time data, and provide precise decision-making support for agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 A schematic diagram of a soil fertility intelligent assessment method and system flow based on image recognition provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of an intelligent soil fertility assessment system based on image recognition provided in an embodiment of the present application.

[0021] Explanation of the reference numerals: image acquisition module 11 , image preprocessing module 12 , feature extraction and analysis module 13 , multi-source information fusion module 14 , soil fertility assessment module 15 , visualization module 16 . DETAILED DESCRIPTION

[0022] The present application provides a soil fertility intelligent assessment method and system based on image recognition, aiming to solve the problems that the sampling range of the ordinary sample method in the prior art is limited, resulting in a one-sided result of soil fertility assessment that cannot accurately reflect the soil condition of the entire plot, and that in order to solve the problem of dynamic changes in soil fertility, sampling and analysis at multiple time points are required, which is not only time-consuming and labor-intensive, but also increases manpower and economic costs. In addition, when the plots are far apart and sampling is inconvenient, the existing methods have low working efficiency and are difficult to meet actual needs. The present application provides a soil fertility intelligent assessment method and system based on image recognition, so that the system proposed in the present application not only overcomes the defects of the prior art in terms of sample limitations, time costs and data dependence, but also improves the efficiency and accuracy of soil fertility assessment through intelligent means, and has broad application prospects.

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0024] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0025] Embodiment 1, as Figure 1 As shown, the present application provides a soil fertility intelligent assessment method and system based on image recognition, the method comprising: Step S100: determining the area to be evaluated; In the embodiment of the present application, the area to be evaluated is an area where soil fertility evaluation needs to be performed.

[0026] Step S200: obtaining land image information of the area to be evaluated, wherein the land image information includes a multispectral image, a thermal imaging image and an RGB image of the area to be evaluated; In the embodiment of the present application, after the area to be evaluated is determined, the land image information of the area to be evaluated can be obtained by the following image acquisition technical means: Drones: Drones can be equipped with multispectral, hyperspectral, infrared and thermal imaging cameras, which can obtain high-resolution images flexibly and at low cost, and are suitable for soil monitoring in large areas such as farmland.

[0027] Satellite remote sensing: Using satellites to obtain multispectral or hyperspectral images is suitable for soil monitoring in large areas. The resolution may be lower than that of drones, but the coverage is wider.

[0028] The types of images collected include: Multispectral imagery: Contains multiple bands (such as red, green, blue, near infrared, shortwave infrared, etc.) and is used to analyze vegetation health, moisture content, and soil properties.

[0029] Thermal imaging images: record the thermal radiation information of soil and vegetation, and can reflect the temperature changes and moisture content of the soil.

[0030] RGB image: A standard red, green, and blue visible light image, commonly used for preliminary visual assessment and simple color analysis.

[0031] Step S300: performing image preprocessing on the land image information to remove various interference factors; In the embodiment of the present application, the detailed steps of image preprocessing include: Image correction, the image correction comprising: Geometric Correction: Ensure that RGB, multispectral and thermal images are accurately aligned in space. Use Ground Control Points (GCPs) or Digital Elevation Models (DEMs) for geometric correction.

[0032] Atmospheric correction (applicable to multispectral images): eliminates the effects of atmospheric scattering and absorption to ensure the accuracy of spectral data.

[0033] Radiometric correction (applicable to multispectral and thermal imaging images): converting the raw data recorded by the sensor into physical quantities (such as reflectance and brightness temperature).

[0034] The purpose of image preprocessing is to ensure that all image data are consistent in geometric position, spectral response and radiation intensity, laying the foundation for subsequent fusion and analysis.

[0035] Image registration, the image registration comprising: Feature point registration: Feature point-based registration: detect and match feature points (such as corners, edges) in RGB, multispectral and thermal images. Common algorithms include SIFT (Scale Invariant Feature Transform) and SURF (Speeded Up Robust Features).

[0036] Image block matching, the image block matching includes: Block matching based registration: The image is divided into multiple blocks and the alignment of different images is adjusted by matching the similarities of these blocks.

[0037] This ensures that RGB, multispectral and thermal images are aligned in the same coordinate system, allowing information from different sources to be fused and analyzed at the pixel level; Step S400: performing vegetation coverage and health assessment based on the multispectral image: calculating the vegetation index of the area to be assessed through the multispectral image, and inferring the vegetation health status and soil organic matter content of the area to be assessed based on the vegetation index of the area; In the embodiment of the present application: it is through obtaining a multispectral image of the area to be evaluated to provide the red light, near infrared and blue light band data required for calculating NDVI and EVI.

[0038] The preparation conditions before the analysis are as follows: Use a multispectral camera or satellite remote sensing technology to collect multispectral images of the area to be evaluated. Ensure that the image quality is high, contains accurate red, near-infrared and blue light band information, and perform necessary calibration and geometric correction on the image to eliminate distortion and errors that may occur during the imaging process.

[0039] In a possible implementation, step S400 further includes: Step S410, calculating the NDVI value of each pixel point according to the red light and near infrared bands in the multispectral image to generate an NDVI image; Step S420, calculating the EVI value of each pixel according to the red light, near infrared light and blue light bands to generate an EVI image; Step S430, evaluating the vegetation coverage of the soil area according to the NDVI; Step S440, further distinguishing the health status of the high-density vegetation area according to the EVI, and identifying whether the vegetation growth is affected by insufficient water or nutrients; Step S450, inferring the vegetation health status and soil organic matter content of the area to be evaluated based on the vegetation coverage and health status of the soil area evaluated by the NDVI and EVI; Specifically: In step S410, NDVI (Normalized Difference Vegetation Index) is calculated to evaluate the vegetation coverage of the soil area; Formula: NDVI = (NIR - RED) / (NIR + RED); NIR: reflectance in the near-infrared band; RED: reflectance of red light band; The operation details are as follows: Pixel reading: Traverse each pixel in the multispectral image and read the red band (RED) and near infrared band (NIR) reflectance values ​​of the pixel.

[0040] NDVI calculation: For each pixel, substitute the formula to calculate and generate the corresponding NDVI value. The NDVI value range is usually between -1 and 1. The closer the value is to 1, the better the vegetation coverage.

[0041] NDVI image generation: The NDVI values ​​of all pixels are aggregated to generate an NDVI image. The brightness or color of different areas in the image represents the status of vegetation coverage.

[0042] In step S420, the Enhanced Vegetation Index (EVI) is calculated to further distinguish the health status of high-density vegetation areas and identify whether vegetation growth is affected by insufficient water or nutrients.

[0043] Formula: EVI = G * (NIR - RED) / (NIR + C1 * RED - C2 * BLUE + L); NIR: reflectance in the near-infrared band; RED: reflectance of red light band; BLUE: reflectance of blue light band; G: gain factor (usually 2.5); C1 and C2: correction coefficients (usually 6 and 7.5); L: Adjustment item (usually 1).

[0044] The operation details are as follows: Pixel reading: Traverse each pixel in the multispectral image and read the reflectance values of the red (RED), near-infrared (NIR), and blue (BLUE) bands of that pixel.

[0045] EVI calculation: For each pixel, substitute into the formula for calculation to generate the corresponding EVI value. The EVI value can more effectively evaluate high-density vegetation areas, especially in the case of strong red light reflection.

[0046] EVI image generation:汇总 the EVI values of all pixels to generate an EVI image. This image helps identify whether vegetation growth is affected by environmental factors (such as insufficient water or nutrients).

[0047] In step S430, the vegetation coverage assessment is to evaluate the vegetation coverage of the soil area based on the NDVI image.

[0048] The specific operations are as follows: Threshold setting: Set the threshold range of the NDVI value according to experience or historical data, for example: NDVI > 0.6 indicates high vegetation coverage; 0.2 < NDVI ≤ 0.6 indicates medium vegetation coverage; NDVI ≤ 0.2 indicates low vegetation coverage or no vegetation; Region division: Use these thresholds to partition the NDVI image and mark the high-coverage, medium-coverage, and low-coverage regions.

[0049] Result analysis: Evaluate the overall vegetation coverage of the soil area and identify areas that need to be monitored or improved.

[0050] In step S440, the vegetation health status assessment is to further analyze the health status of the high-density vegetation area based on the EVI image and identify whether vegetation growth is affected by insufficient water or nutrients.

[0051] The specific operations are as follows: EVI value analysis: Analyze the EVI values of different regions in the EVI image and identify regions with low EVI values, which may mean that vegetation growth is affected by environmental stress (such as insufficient water or nutrients).

[0052] Health status classification: Divide the regions with low EVI values into healthy, sub-healthy, and unhealthy vegetation regions for further measures.

[0053] In step S450, the soil organic matter content is speculated by analyzing the high and low of NDVI and EVI, speculating the vegetation health status of the soil area, and indirectly speculating the soil organic matter content.

[0054] The specific operations are as follows: It should be noted that there is a misspelling in the original text "汇总" which should be "汇总". The corrected translation is provided above. Health and organic matter relationship: High NDVI and EVI values ​​are generally associated with soils with high organic matter content, because healthy vegetation tends to grow in soils rich in organic matter.

[0055] Estimated organic matter content: Based on the comprehensive situation of NDVI and EVI in the area, the soil organic matter content is estimated.

[0056] Results summary: The organic matter content of the soil area was evaluated overall by combining the analysis results of NDVI and EVI, and the results were integrated into the soil fertility assessment model.

[0057] Through the above steps, this embodiment uses multispectral images to calculate NDVI and EVI, which can effectively evaluate the vegetation coverage and health status of the soil area and infer the soil organic matter content. This series of operations lays a data foundation for subsequent soil fertility assessment and provides a scientific basis for soil management decisions.

[0058] Step S500: inferring soil characteristics based on the RGB image: extracting the texture features of the soil surface through the RGB image, analyzing the arrangement, cracks and pore structure of soil particles, so as to infer the stability and porosity of the soil structure; In the embodiment of the present application: it is to obtain a high-resolution RGB image of the area to be evaluated, to extract the texture features of the soil surface, and to analyze the arrangement, cracks and pore structure of the soil particles.

[0059] The preparation conditions before the analysis are as follows: Image acquisition: Use high-definition video equipment or drones to collect RGB images of the area to be assessed under appropriate lighting conditions. Make sure the image resolution is high enough to clearly capture the details of the soil surface.

[0060] Image preprocessing: Perform basic preprocessing on the collected RGB images, including image denoising, brightness and contrast adjustment, to ensure that the details in the image are fully displayed.

[0061] It obtains multispectral images of the area to be assessed to provide the red, near-infrared and blue light band data required to calculate NDVI and EVI.

[0062] In a possible implementation, step S500 further includes: Step S510, extracting texture features based on gray level co-occurrence matrix (GLCM) from the RGB image; Step S520, using the LBP method to extract the local texture pattern of the image, and representing the LBP features in the form of a histogram; Step S530, performing frequency decomposition on the image using wavelet transform to extract texture features at different scales; Step S540, the soil structure estimation result is obtained by combining the analysis results of GLCM, LBP and wavelet transform.

[0063] Specifically: In step S510, the texture feature extraction based on the gray level co-occurrence matrix (GLCM) is to extract the texture features based on the gray level co-occurrence matrix (GLCM) from the RGB image, which is used to analyze the microstructure of the soil surface; The specific operations are as follows: Grayscale processing: Convert RGB images into grayscale images to simplify subsequent texture analysis.

[0064] GLCM calculation: Select a suitable pixel offset and direction to calculate the gray level co-occurrence matrix. GLCM is used to describe the spatial relationship between pixel gray values. Commonly used offset directions include horizontal, vertical and diagonal directions.

[0065] Texture feature extraction: A series of texture features such as energy, contrast, entropy and homogeneity are extracted from GLCM. These features can reflect the roughness, particle arrangement and distribution of cracks on the soil surface.

[0066] Feature analysis: Analyze the extracted texture features to preliminarily determine the structure and porosity of the soil. For example, high contrast may indicate that the soil surface particles are rough and uneven, while high uniformity may indicate that the soil structure is relatively stable.

[0067] In step S520, extracting local texture using local binary pattern (LBP) is to extract local texture pattern of the image by LBP method to further analyze the subtle features of the soil surface; The specific operations are as follows: LBP calculation: For each pixel in the grayscale image, the pixel value in its neighborhood is compared with the pixel value to generate a binary pattern. The LBP value is usually expressed by converting the binary pattern into a decimal value.

[0068] Histogram construction: The LBP feature is represented in the form of a histogram, and the frequency distribution of different LBP values ​​in the image is statistically analyzed. The histogram can reflect the local texture distribution in the image.

[0069] Feature analysis: By analyzing the distribution of LBP histograms, the local structural characteristics of the soil surface can be determined. For example, a highly concentrated histogram may indicate that the soil surface texture is relatively consistent, while a dispersed histogram may indicate that the soil surface structure is complex.

[0070] In step S530, the frequency decomposition and texture feature extraction using wavelet transform is to perform frequency decomposition on the image using wavelet transform and extract texture features at different scales to capture multi-scale information of soil structure; The specific operations are as follows: Wavelet decomposition: Perform wavelet decomposition on grayscale images to decompose the image into sub-bands of different scales and directions. Commonly used wavelet transforms include discrete wavelet transform (DWT), which can decompose images into low-frequency and high-frequency components.

[0071] Feature extraction: Extract texture features from each sub-band, such as energy distribution, edge features, etc. Low-frequency components usually reflect the overall structure of the image, while high-frequency components reveal details and edge information.

[0072] Multi-scale analysis: By analyzing the texture features at different scales, we can have a more comprehensive understanding of the structural characteristics of the soil surface. The smooth characteristics of the low-frequency components can reflect the overall stability of the soil, while the complex characteristics of the high-frequency components may be related to the roughness and cracks of the soil surface; The multi-feature fusion and soil structure inference in step S540 is to make a more comprehensive inference of soil structure by combining the analysis results of GLCM, LBP and wavelet transform; The specific operations are as follows: Feature integration: The features extracted from GLCM, LBP and wavelet transform are integrated to form a comprehensive feature vector. This vector contains multi-dimensional information of the soil surface, including spatial texture, local pattern and multi-scale frequency characteristics.

[0073] Feature analysis: Analyze the comprehensive feature vector to identify the key features of soil structure. For example, by analyzing the weights of each feature dimension, it is possible to identify which features are most important for predicting soil structural stability and porosity.

[0074] Soil structure inference: Based on the results of feature analysis, the structural stability and porosity of the soil are inferred. For example, if the high-contrast and high-frequency component features in the feature vector account for a high proportion, it may indicate that the soil structure is loose and the porosity is high; while if the low-frequency component and high uniformity features account for a high proportion, it may indicate that the soil structure is compact and stable.

[0075] This embodiment can analyze the texture structure of the soil surface in a multi-level and multi-dimensional manner through the GLCM, LBP and wavelet transform features extracted from the RGB image. Combining these features can effectively infer the structural stability and porosity of the soil, thereby providing a key basis for further evaluation of soil fertility; Step S600: infer the moisture change of the soil surface according to the temperature distribution of the thermal imaging image, and identify the drought area and the area with sufficient moisture; In the embodiment of the present application: it is to obtain a thermal imaging image of the area to be evaluated to analyze the temperature distribution of the soil surface, thereby inferring the moisture change of the soil; The preparation conditions before the analysis are as follows: Image acquisition: Use thermal imaging equipment or drones equipped with thermal imaging sensors to collect thermal imaging images of the area to be assessed at a suitable time period (usually noon or afternoon on a sunny day). Ensure that the resolution and coverage of the image can fully reflect the temperature distribution of the soil.

[0076] Image preprocessing: Preprocess the collected thermal imaging images, including correcting the temperature value, removing noise, and adjusting the contrast and brightness of the image to make the temperature information clearer.

[0077] In a possible implementation, step S600 further includes: Step S610, analyzing the temperature distribution in the thermal imaging image to determine the temperature change pattern of the soil surface; Step S620, inferring the moisture change of the soil surface according to the temperature distribution, and identifying the drought area and the area with sufficient moisture; Step S630, identifying drought areas and water-sufficient areas in the soil to further analyze soil fertility and plant growth conditions; Step S630, integrating the results of temperature distribution and humidity analysis to provide a basis for soil fertility assessment.

[0078] Specifically: In step S610, the temperature distribution analysis is to determine the temperature variation pattern of the soil surface by analyzing the temperature distribution in the thermal imaging image; The specific operations are as follows: Temperature mapping: Map the temperature data in the thermal imaging image to a pseudo-color image, with different colors representing different temperature ranges. Usually blue is used to represent low temperature areas, and red is used to represent high temperature areas.

[0079] Temperature gradient calculation: By calculating the temperature gradient in the image, the temperature change pattern can be identified. The temperature gradient can help identify local temperature differences and infer changes in soil moisture.

[0080] Temperature hotspot identification: Identify the areas with the highest and lowest temperatures in the image, marking possible drought areas and areas with sufficient water. Areas with high temperatures generally indicate low soil moisture, while areas with low temperatures generally indicate high soil moisture.

[0081] In step S620, soil moisture estimation is to estimate the moisture change of the soil surface based on the temperature distribution, and identify the drought area and the area with sufficient moisture; The specific operations are as follows: Relationship between temperature and humidity: The soil humidity is estimated by using the relationship model between soil temperature and humidity and combining the collected temperature data. Generally speaking, the temperature of moist soil changes slowly, so it appears as a lower temperature in the thermal imaging image; the temperature of dry soil changes quickly, which appears as a higher temperature.

[0082] Humidity distribution map generation: Generate soil moisture distribution map based on the estimated humidity value. Use different colors or grayscale to represent areas with different humidity, such as blue for wet areas and red for dry areas.

[0083] Humidity change pattern recognition: Analyze the humidity distribution map and identify obvious humidity change patterns. For example, if there is a large area of ​​high temperature and it overlaps with the area of ​​low humidity in the same field, the area may be at risk of drought.

[0084] In step S630, the identification of drought and water-sufficient areas is performed by identifying drought areas and water-sufficient areas in the soil, so as to further analyze the soil fertility and plant growth conditions; The specific operations are as follows: Region division: Based on the moisture distribution map, the soil region is divided into arid region, well-watered region, and transitional region. The boundaries of each region are determined using a threshold method. For example, a moisture threshold can be set, and regions above the threshold are marked as well-watered, and regions below the threshold are marked as arid.

[0085] Drought risk assessment: Further analysis of drought areas to assess their potential impact on soil fertility and plant growth. For example, if an area is identified as drought several times in a row, irrigation or soil improvement measures may be needed.

[0086] Well-Watered Area Analysis: Well-watered areas are analyzed to determine if they are at risk of becoming overly wet, which could lead to poor soil structure or oxygen deprivation for plant roots.

[0087] In step S640, the comprehensive humidity analysis and soil fertility assessment is to provide a basis for soil fertility assessment by integrating the results of temperature distribution and humidity analysis; The specific operations are as follows: Data integration: Integrate the temperature information in the thermal imaging image with the humidity distribution map to generate a comprehensive humidity analysis map. This map can be combined with other data (such as vegetation index, RGB image analysis results) for comprehensive evaluation.

[0088] Analysis of the impact of humidity on fertility: By comprehensively analyzing the impact of soil humidity on fertility, a preliminary assessment of soil fertility is obtained. Moderately moist soils usually have higher fertility, while soils that are too dry or too wet may be at risk of insufficient fertility.

[0089] Adjustments and Optimization: Based on the analysis results, adjustments are recommended for specific areas, such as optimizing soil moisture through irrigation or drainage measures to improve soil fertility.

[0090] This embodiment analyzes the temperature distribution of the soil surface through thermal imaging images, which can effectively infer the moisture changes of the soil and identify drought and water-sufficient areas. This information is crucial for soil fertility assessment and can provide a scientific basis for subsequent soil improvement and plant growth management.

[0091] Step S700: Integrate the high-resolution texture features of the RGB image, the spectral information of the multispectral image, and the inferred results of the temperature distribution of the thermal imaging image to generate a comprehensive analysis result; then classify the soil type and evaluate the fertility level based on the comprehensive analysis result.

[0092] In a possible implementation, step S700 further includes: Step S710, weighted voting is performed on the independent analysis results of the RGB image, the multispectral image, and the thermal imaging image, and each result is assigned a different weight according to its credibility or importance; Step S720, setting decision rules, integrating the independent analysis results according to the predetermined rules; Step S730, using an integrated learning method such as a decision tree or a random forest, taking the analysis results from different sources as input features, and generating a comprehensive analysis result.

[0093] Step S740, using soil sample images of known fertility levels to annotate the model and generate a training data set; Step S750, using a machine learning or deep learning model to perform classification or regression analysis on the comprehensive analysis results to predict soil fertility.

[0094] Specifically: In step S710, the independent analysis results are obtained by respectively obtaining high-resolution texture feature analysis of RGB images, spectral information analysis of multispectral images, and temperature distribution inference results of thermal imaging images, so as to provide basic data for subsequent comprehensive analysis; The specific operations are as follows: RGB image analysis: Texture feature extraction of RGB images is performed using local binary patterns (LBP), gray-level co-occurrence matrix (GLCM) and wavelet transform. The extracted features include the arrangement of soil particles, cracks and pore structures, etc., which are used to infer the stability and porosity of soil structure.

[0095] Multispectral image analysis: Calculate vegetation indices (such as NDVI and EVI) based on the red and near-infrared bands of multispectral images to assess vegetation coverage and health and infer soil organic matter content.

[0096] Thermal imaging image analysis: Analyze the temperature distribution in thermal imaging images, infer changes in soil surface moisture, and identify drought areas and areas with sufficient water.

[0097] The goal of weighted voting for independent analysis results is to weight each analysis result according to its importance and credibility to ensure the accuracy of the comprehensive analysis; The specific operations are as follows: Operation details: Weighting: Different weights are assigned based on the importance or historical performance of the analysis results. For example, the vegetation index from a multispectral image may have a higher weight for inferring organic matter content, while the temperature distribution from a thermal image may have a higher weight for humidity changes. These weights can be determined through expert experience or historical data.

[0098] Weighted voting: Each analysis result is multiplied by its corresponding weight, and then the weighted results are weighted voted. If the voting results of a certain category (such as soil type) exceed the preset threshold, then this category is selected as the final comprehensive analysis result.

[0099] In step S720, decision rule setting and integration is to effectively integrate the analysis results of RGB, multi-spectral and thermal imaging images by setting comprehensive decision rules to generate a final comprehensive analysis result; The specific operations are as follows: Rule setting: Based on the specific soil type and fertility assessment requirements, set the corresponding decision rules. For example, if the RGB image shows high porosity and the multispectral image shows high organic matter content, the soil type is high fertility soil; if the thermal image shows too high temperature and low humidity, it may be drought soil.

[0100] Decision integration: Based on the set decision rules, the individual analysis results are logically combined to generate a comprehensive classification result. For example, if the RGB analysis results and the multispectral analysis results both indicate that the soil is healthy, but the thermal imaging analysis shows local drought, the comprehensive analysis result may be "healthy but attention should be paid to water management."

[0101] In step S730, the comprehensive analysis result is generated by using an integrated learning method such as a decision tree or a random forest, using analysis results from different sources as input features to generate a final comprehensive analysis result.

[0102] The specific operation is as follows: Feature input: The texture features of the RGB image, the vegetation index of the multispectral image, and the temperature features of the thermal image are used as input features and input into the decision tree or random forest model. Each set of data represents the multidimensional features of a soil sample.

[0103] Model training: Use historical data labeled with soil type and fertility level to train a decision tree or random forest model. The model learns the relationship between different features and soil type and fertility level to form decision rules.

[0104] Result output: The analysis results of the area to be evaluated are input into the trained model to generate comprehensive analysis results.

[0105] In step S740, a training data set is generated by constructing a data set for training a machine learning or deep learning model, wherein each data sample contains an image feature and a known soil fertility grade label.

[0106] The specific operations are as follows: Sample image collection: Collect images of soil samples of known fertility levels. These images can be a combination of RGB, multispectral, and thermal images, ensuring that the samples cover a variety of soil types and fertility levels.

[0107] Feature extraction: Extract texture features (such as GLCM, LBP, wavelet transform features) from RGB images.

[0108] Calculate vegetation indices (NDVI, EVI) from multispectral images and extract related spectral features.

[0109] Temperature distribution features are extracted from thermal imaging images to infer soil moisture.

[0110] Data annotation: Add known fertility level labels to each image sample. These labels can come from field measurements or historical records, and are usually divided into high, medium, and low levels.

[0111] Dataset division: Divide the dataset into training set, validation set and test set. The training set is used to train the model, the validation set is used to adjust the model parameters, and the test set is used to evaluate the performance of the model.

[0112] In step S750, a suitable machine learning or deep learning model may be selected according to data characteristics and task requirements to be used for classification or regression analysis; The specific operations are as follows: Step S751, model type selection: Machine learning model: You can choose classic algorithms such as random forest, support vector machine (SVM), k-nearest neighbor algorithm (KNN), etc.

[0113] Deep learning model: You can choose convolutional neural network (CNN) to process image data, or more complex network architectures such as ResNet, Inception, etc.

[0114] Model initialization: Initialize the model parameters, determine the network structure (such as the number of layers and nodes) or the hyperparameters of the machine learning model (such as the number of trees, depth, etc.).

[0115] Step S752, model training, aims to adjust the model parameters through training data so that it can effectively learn the relationship between image features and soil fertility levels.

[0116] The specific operations are as follows: Feature input: Input the features extracted from the image into the model. For deep learning models, you can directly input the original image and let the model automatically extract features.

[0117] Loss function selection: Choose an appropriate loss function based on the task.

[0118] Classification task: commonly used cross entropy loss function.

[0119] Regression task: commonly used mean square error (MSE) loss function.

[0120] Back propagation and optimization: Use the back propagation algorithm to calculate the gradient of the loss function and adjust the model parameters. Common optimizers include Adam, SGD, etc.

[0121] Iterative training: Update the model parameters in multiple iterations (epochs), evaluate the performance of the model using the validation set after each iteration, and adjust the hyperparameters to improve the accuracy and robustness of the model.

[0122] Step S753, model verification and testing, is intended to evaluate the performance of the model through the verification set and the test set to ensure that the model can effectively predict soil fertility on unseen data.

[0123] The specific operations are as follows: Validation set evaluation: After each training cycle, the model is evaluated using the validation set to check whether the model is overfitting or underfitting.

[0124] Model tuning: Adjust the model structure or hyperparameters (such as learning rate, regularization coefficient, etc.) based on the results of the validation set.

[0125] Test set evaluation: After training is completed, the model is finally evaluated using the test set to calculate the model's accuracy, recall, F1 score and other indicators.

[0126] Cross-validation: Methods such as k-fold cross-validation can be used to further verify the stability and generalization ability of the model.

[0127] Step S754, classification or regression prediction of comprehensive analysis results: Use the trained model to classify or regress the comprehensive analysis results of the new area to generate soil fertility assessment results.

[0128] The specific operations are as follows: New data feature extraction: Preprocess and extract features from the image data of the new area to generate feature vectors similar to the training data.

[0129] Model predictions: Classification prediction: The model inputs the characteristics of new data and outputs the corresponding soil fertility level (such as high, medium, and low).

[0130] Regression prediction: If it is a regression task, the model will output a continuous fertility value, indicating the fertility level of the soil.

[0131] Interpretation of results: Interpret the output of the model and map it into practical soil management recommendations, such as fertilizer adjustment, irrigation plan formulation, etc.

[0132] Step S755, result feedback and model update, based on the feedback from actual applications, continuously update and improve the model to improve the accuracy and reliability of soil fertility prediction.

[0133] The specific operations are as follows: Results verification: The model's predictions were compared with actual soil fertility measurements to assess the accuracy of the predictions.

[0134] Model update: If you find that the model makes inaccurate predictions in certain situations, you can retrain the model with new sample data or adjust the model's structure and parameters.

[0135] Continuous optimization: Regularly update the dataset and retrain the model to ensure its robustness and accuracy in different seasons and climate conditions.

[0136] Finally, it can be presented visually: soil types and fertility levels can be visualized in the form of maps or tables, so that growers and decision makers can intuitively understand the soil conditions and formulate corresponding agricultural management measures accordingly.

[0137] Embodiment 2, based on the same inventive concept as the soil fertility intelligent assessment method and system based on image recognition in the above-mentioned embodiment, Figure 2 As shown, the present application provides a soil fertility intelligent assessment system based on image recognition, and the system and method embodiments in the present application embodiments are based on the same inventive concept. The system includes: Image acquisition module 11: The image acquisition module is responsible for acquiring land images of the area to be evaluated, including RGB images, multispectral images and thermal imaging images, and transmitting the acquired image data to the image preprocessing module for processing; Image preprocessing module 12: preprocesses the collected images, including denoising, correction, image registration and resolution adjustment, and transmits the processed image data to the feature extraction and analysis module as input for feature analysis; Feature extraction and analysis module 13: extract key features from different types of images and analyze them, and transmit the analysis results to the multi-source information fusion module; RGB images: extract texture features and analyze soil structure stability and porosity; Multispectral images: calculate vegetation index, assess vegetation coverage and health, and infer soil organic matter content and nutrient status; Thermal imaging: Analyze temperature distribution, infer soil moisture, and identify drought and well-watered areas; Multi-source information fusion module 14: fuses feature data from different image sources to form a comprehensive soil feature description; Specifically, weighted voting, rule integration or integrated learning model is used to combine the independent analysis results of RGB images, multispectral images and thermal imaging images to generate comprehensive analysis results, and the fused comprehensive analysis results are transmitted to the soil fertility assessment module; Soil Fertility Assessment Module 15: Based on the integrated analysis results, soil type classification and fertility level assessment are carried out.

[0138] Specifically, through machine learning or deep learning: use the trained model to perform classification or regression analysis on the comprehensive analysis results to predict the soil fertility level.

[0139] Furthermore, the system is also used to implement the following functions: The image preprocessing methods in the image preprocessing module include: Noise removal: Use Gaussian filtering or median filtering to remove noise from the image to ensure that the extracted features are not interfered by noise.

[0140] Enhance contrast: Use histogram equalization or adaptive histogram equalization to increase the contrast of the image and make the texture features more obvious.

[0141] Image registration: spatially align images from different sources to ensure the accuracy of subsequent feature fusion.

[0142] Furthermore, the system is also used to implement the following functions: The analysis method of multispectral images in the feature extraction and analysis module is as follows: According to the red and near-infrared bands in the multispectral image, the NDVI value of each pixel is calculated to generate an NDVI image; According to the red, near-infrared and blue light bands, the EVI value of each pixel is calculated to generate an EVI image; Based on NDVI, assess the vegetation coverage of the soil area; The EVI can be used to further differentiate the health of high-density vegetation areas and identify whether vegetation growth is affected by water or nutrient deficiency; Based on the vegetation coverage and health status of the soil area assessed by the levels of NDVI and EVI, the vegetation health status and soil organic matter content of the area to be assessed are inferred.

[0143] Furthermore, the system is also used to implement the following functions: The analysis method of RGB images in the feature extraction and analysis module is as follows: Extract texture features based on gray-level co-occurrence matrix (GLCM) from RGB images; The local texture pattern of the image is extracted using the LBP method, and the LBP features are represented in the form of a histogram; Use wavelet transform to perform frequency decomposition on the image and extract texture features at different scales; By combining the analysis results of GLCM, LBP and wavelet transform, the soil structure inference results are obtained.

[0144] Furthermore, the system is also used to implement the following functions: The method for generating comprehensive analysis results in the multi-source information fusion module is as follows: Perform weighted voting on the independent analysis results of RGB images, multispectral images, and thermal images, with each result being assigned a different weight based on its credibility or importance; Set decision rules and integrate the results of each independent analysis according to the predetermined rules; Use ensemble learning methods such as decision trees or random forests to take analysis results from different sources as input features and generate a comprehensive analysis result.

[0145] Furthermore, the system is also used to implement the following functions: In the soil fertility assessment module, the method for classifying soil types and assessing fertility levels based on comprehensive analysis results is as follows: Use soil sample images with known fertility levels to annotate the model and generate a training dataset; Use machine learning or deep learning models to perform classification or regression analysis on the comprehensive analysis results to predict soil fertility.

[0146] Furthermore, the system is also used to implement the following functions: The system also includes a visualization module, which can use visualization tools to generate various charts to intuitively display the characteristics and fertility status of the soil; The chart types are as follows: Map visualization: Use heat maps or contour maps to display information such as soil fertility levels and moisture distribution in different areas. Use color depth or height changes to indicate different fertility levels and moisture values.

[0147] Time series graph: Shows the changing trend of soil fertility or moisture in a specific area over time, helping users understand long-term changes.

[0148] Bar charts and pie charts: used to show the distribution ratio of different soil types or the distribution of fertility levels in different areas.

[0149] Scatter plot: shows the correlation between different features (such as NDVI, EVI and fertility level) to help identify key factors; It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0150] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0151] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. An intelligent soil fertility assessment method based on image recognition, characterized in that: The method comprises: Identify the area to be assessed; Obtaining land image information of the area to be evaluated, wherein the land image information includes a multispectral image, a thermal imaging image, and an RGB image of the area to be evaluated; Performing image preprocessing on the land image information to remove various interference factors; Performing vegetation coverage and health assessment based on the multispectral image: calculating the vegetation index of the area to be assessed through the multispectral image, and inferring the vegetation health status and soil organic matter content of the area to be assessed based on the vegetation index of the area; Inferring soil characteristics based on the RGB image: extracting texture features of the soil surface through the RGB image, analyzing the arrangement, cracks and pore structure of soil particles, so as to infer the stability and porosity of the soil structure; Based on the temperature distribution of thermal imaging images, the moisture changes of the soil surface can be inferred to identify drought areas and areas with sufficient moisture; The high-resolution texture features of RGB images, the spectral information of multispectral images, and the inferred results of temperature distribution of thermal imaging images are integrated to generate comprehensive analysis results; then, soil types are classified and fertility levels are evaluated based on the comprehensive analysis results.

2. The soil fertility intelligent assessment method based on image recognition according to claim 1, characterized in that: Performing image preprocessing on the land image information to remove various interference factors, the method further includes: Noise removal: Use Gaussian filtering or median filtering to remove noise from the image to ensure that the extracted features are not affected by noise; Enhance contrast: Use histogram equalization or adaptive histogram equalization to improve the contrast of the image and make the texture features more obvious; Image registration: spatially align images from different sources to ensure the accuracy of subsequent feature fusion.

3. The soil fertility intelligent assessment method based on image recognition according to claim 1, characterized in that; Performing vegetation coverage and health assessment based on the multispectral image: calculating the vegetation index of the area to be assessed through the multispectral image, and inferring the vegetation health status and soil organic matter content of the area to be assessed based on the vegetation index of the area, the method comprising: According to the red and near-infrared bands in the multispectral image, the NDVI value of each pixel is calculated to generate an NDVI image; According to the red, near-infrared and blue light bands, the EVI value of each pixel is calculated to generate an EVI image; Based on NDVI, assess the vegetation coverage of the soil area; The EVI can be used to further differentiate the health of high-density vegetation areas and identify whether vegetation growth is affected by water or nutrient deficiency; Based on the vegetation coverage and health status of the soil area assessed by the levels of NDVI and EVI, the vegetation health status and soil organic matter content of the area to be assessed are inferred.

4. The soil fertility intelligent assessment method based on image recognition according to claim 1, characterized in that: Inferring soil characteristics based on the RGB image: extracting soil surface texture features through the RGB image, analyzing the arrangement, cracks and pore structure of soil particles, and inferring the stability and porosity of the soil structure, the method comprising: Extract texture features based on gray-level co-occurrence matrix (GLCM) from RGB images; The local texture pattern of the image is extracted using the LBP method, and the LBP features are represented in the form of a histogram; Use wavelet transform to perform frequency decomposition on the image and extract texture features at different scales; By combining the analysis results of GLCM, LBP and wavelet transform, the soil structure inference results are obtained.

5. The soil fertility intelligent assessment method based on image recognition according to claim 1, characterized in that: The high-resolution texture features of the RGB image, the spectral information of the multispectral image, and the inferred results of the temperature distribution of the thermal imaging image are integrated to generate a comprehensive analysis result. The method further includes: Perform weighted voting on the independent analysis results of RGB images, multispectral images, and thermal images, with each result being assigned a different weight based on its credibility or importance; Set decision rules and integrate the results of each independent analysis according to the predetermined rules; The analysis results from different sources are used as input features to generate a comprehensive analysis result.

6. The soil fertility intelligent assessment method based on image recognition according to claim 5, characterized in that: The method of using the analysis results from different sources as input features to generate a comprehensive analysis result is as follows: Perform weighted voting on the independent analysis results of RGB images, multispectral images, and thermal images, with each result being assigned a different weight based on its credibility or importance; Set decision rules and integrate the results of each independent analysis according to the predetermined rules; Use ensemble learning methods such as decision trees or random forests to take analysis results from different sources as input features and generate a comprehensive analysis result.

7. The method for intelligent soil fertility assessment based on image recognition according to claim 1, characterized in that: The method of classifying soil types and evaluating soil fertility levels according to the comprehensive analysis results comprises: Use soil sample images with known fertility levels to annotate the model and generate a training dataset; The comprehensive analysis results are classified or regressed to predict soil fertility.

8. An intelligent soil fertility assessment system based on image recognition, the system being used to execute the method according to any one of claims 1 to 7, characterized in that: The system comprises: Image acquisition module: The image acquisition module is responsible for acquiring land images of the area to be evaluated, including RGB images, multispectral images and thermal imaging images, and transmitting the acquired image data to the image preprocessing module for processing; Image preprocessing module: preprocesses the collected images, including denoising, correction, image registration and resolution adjustment, and transmits the processed image data to the feature extraction and analysis module as input for feature analysis; Feature extraction and analysis module: extracts key features from different types of images and analyzes them, and transmits the analysis results to the multi-source information fusion module; RGB images: extract texture features and analyze soil structure stability and porosity; Multispectral images: calculate vegetation index, assess vegetation coverage and health, and infer soil organic matter content and nutrient status; Thermal imaging: Analyze temperature distribution, infer soil moisture, and identify drought and well-watered areas; Multi-source information fusion module: fuses feature data from different image sources to form a comprehensive soil feature description; Specifically, weighted voting, rule integration or integrated learning model is used to combine the independent analysis results of RGB images, multispectral images and thermal imaging images to generate comprehensive analysis results, and the fused comprehensive analysis results are transmitted to the soil fertility assessment module; Soil Fertility Assessment Module: Based on the integrated analysis results, soil type classification and fertility level assessment are carried out; Specifically, through machine learning or deep learning: use the trained model to perform classification or regression analysis on the comprehensive analysis results to predict the soil fertility level.

Images