Field-level agricultural machine driving efficiency suitability evaluation method and system based on remote sensing image

Abstract

The invention provides a remote sensing image-based in-field agricultural machine driving efficiency suitability evaluation method. The method comprises the steps of S1, extracting fine field attribute information based on a high-resolution remote sensing image; s2, extracting field obstacle information and field terrain information based on the field attribute information; s3, constructing a key evaluation index set influencing the running efficiency of the agricultural machinery in the field; s4, index weighting based on a geographical weighted attention mechanism; s5, establishing an in-field agricultural machine driving efficiency suitability evaluation model, and positioning key negative impact indexes; and S6, grading the suitability level of the agricultural machine driving efficiency based on a natural breakpoint method. The invention further provides a system for evaluating the suitability of the running efficiency of the agricultural machinery in the field based on the remote sensing image. According to the method, the field-level key index set is constructed, the geographical weighted regression and the attention mechanism are innovatively combined, the regional difference and the field characteristics are fully considered, and the scientificity and the practicability of evaluation are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of farmland monitoring technology, and more specifically, to a method and system for evaluating the suitability of field-level agricultural machinery driving efficiency based on remote sensing images. Background Art

[0002] Scientifically and accurately evaluating the suitability of agricultural machinery operating efficiency at the field level not only provides data support for pre-site selection and post-mechanization evaluation, but also promotes the deep integration of farmland, agricultural machinery, and agronomy, driving the intelligent development of agricultural mechanization. In recent years, the integration of high-resolution remote sensing technology and deep learning has provided a rich data foundation for evaluating the suitability of agricultural machinery operating efficiency at the field level. However, a scientific and reasonable evaluation model is also crucial for this purpose. Existing evaluation methods and techniques mostly use fixed weighting schemes, which ignore the integration of macro-regional and micro-field characteristics and, therefore, cannot achieve adaptive and accurate evaluation across a large area. Most research methods fail to fully consider regional spatial differences and exploit the dynamic relationship between geographic information and field characteristics, resulting in a lack of regional specificity in their results. Furthermore, previous studies have focused solely on the high or low levels of agricultural machinery operating efficiency suitability at the field level, failing to identify key negative impact factors and lacking specific guidance for translating abstract scores into concrete transformation goals. Summary of the Invention

[0003] In response to the problems in the background technology, the present invention proposes a field-level agricultural machinery driving efficiency suitability evaluation method that combines a large visual segmentation model, a multi-task learning network, a geographically weighted regression and an attention mechanism.

[0004] The method for evaluating the suitability of agricultural machinery driving efficiency within a field based on remote sensing images of the present invention includes: S1, extracting detailed field attribute information based on high-resolution remote sensing images; S2, extracting field obstacle information and field terrain information based on the field attribute information; S3, constructing a set of key evaluation indicators that affect the driving efficiency of agricultural machinery within a field; S4, weighting indicators based on a geographically weighted attention mechanism; S5, establishing a model for evaluating the suitability of agricultural machinery driving efficiency within a field and locating key negative impact indicators; S6, grading the suitability level of agricultural machinery driving efficiency based on a natural breakpoint method.

[0005] The present invention also provides a corresponding evaluation system for the suitability of agricultural machinery driving efficiency at the field level, which includes a computer executable program capable of completing the method steps described above.

[0006] This paper addresses the difficulty of adaptively and accurately evaluating agricultural machinery operating efficiency across large fields by proposing an intelligent evaluation method that combines refinement with broad adaptability. By constructing a set of key indicators at the field level and innovatively combining geographically weighted regression with an attention mechanism, this method fully considers regional differences and field characteristics, significantly enhancing the scientific and practical nature of the evaluation. This method integrates remote sensing data, adaptively and dynamically adjusts indicator weights, and precisely identifies key influencing factors, providing reliable decision-making support for improving agricultural machinery operating efficiency and precision farmland transformation. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to make the present invention more easily understood, the present invention will be described in more detail with reference to the specific embodiments shown in the accompanying drawings. These drawings only depict typical embodiments of the present invention and should not be considered as limiting the scope of protection of the present invention.

[0008] Figure 1 is a flow chart of the method of the present invention.

[0009] Figure 2 This is the result map of field vector extraction.

[0010] Figure 3 This is the result map of obstacle extraction in the field.

[0011] Figure 4 It is a field topographic map.

[0012] Figure 5 It is a set of evaluation indicators that key influencing the driving efficiency of agricultural machinery in the field.

[0013] Figure 6 Agricultural machinery driving suitability classification map within the field. DETAILED DESCRIPTION

[0014] The following describes the embodiments of the present invention with reference to the accompanying drawings so that those skilled in the art can better understand the present invention and implement it. However, the enumerated embodiments are not intended to limit the present invention. Unless there is a conflict, the following embodiments and the technical features in the embodiments may be combined with each other, wherein the same components are represented by the same figure marks.

[0015] The method of the present invention includes: S1, extracting detailed field attribute information based on high-resolution remote sensing images; S2, extracting field obstacle information and field terrain information based on field boundary information; S3, constructing a set of key evaluation indicators that affect the driving efficiency of agricultural machinery in the field; S4, weighting indicators based on a geographically weighted attention mechanism; S5, establishing a suitability evaluation model for the driving efficiency of agricultural machinery in the field, and locating key negative impact indicators; S6, grading the suitability level of agricultural machinery driving efficiency based on a natural breakpoint method.

[0016] In an embodiment, in step S1, based on the high-resolution remote sensing image of the target area, field attribute information is obtained by using a deep learning field extraction algorithm. In one embodiment, step S1 includes steps S11-S16.

[0017] Preferably, a high spatial resolution remote sensing image of a cloud-free target area is selected and preprocessed (radiometric calibration, atmospheric correction, etc.).

[0018] S11, perform overlapping screening on the masks generated after high-resolution remote sensing image segmentation (preferably using SAM segmentation). Based on the stability score as the evaluation basis, the effective mask list of SAM segmentation is sorted in descending order. The overlap of the masks in the effective mask list is judged in turn. Masks without overlap can be added to a separate field mask list. Masks with overlap are judged based on indicators such as overlap ratio, number of overlapping masks and mask quality. Masks that meet the above field mask judgment conditions are retained, otherwise they are eliminated. The stability score is one of the SAM mask quality indicators, and its formula is as follows:

[0019]

[0020] Preferably, S12, a dilation and corrosion operation is performed on the mask after the first segmentation, and linear non-arable land features such as small roads and ridges are removed from the original image, and the retained unsegmented image area is used as input for the next segmentation.

[0021] Preferably, S13, iterative segmentation is performed, and the high-resolution image is subjected to the initial instance segmentation using a rule-prompted point network of a specified density to generate an initial effective mask. Based on the initial effective mask, an expansion convolution operation is performed to extract the complex area not covered by the mask as an iterative image for iterative segmentation. In each round of iterative segmentation, an overlapping screening mechanism is performed on the effective mask. If the number of iterations of the iterative segmentation is greater than 10 (set as needed), the segmentation is stopped. If the number of iterations is less than 10, it is compared with the previous segmentation result to determine whether the segmentation result is the same. If so, the segmentation is stopped; if the segmentation results are different, the mask coverage is calculated. If the mask coverage is less than 90%, the iterative segmentation is continued. If the coverage is greater than or equal to 90%, the segmentation is stopped.

[0022] Preferably, S14, performs mask stitching of predicted intersection-and-union ratio: 1) vectorizes the field raster mask results obtained by segmentation to achieve separation and independent numbering of each field mask; 2) migrates the predicted intersection-and-union ratio information generated during the segmentation process to the attributes of the vector elements, and sorts all elements according to the information (in descending order); 3) traverses all elements, performs topological relationship checks in turn and adds them to the result list. If the current element has a topological relationship with an existing element in the list, it is removed, thereby forming a vector segmentation result of the field area.

[0023] S15, visualize the vector results of the final field (see Figure 2 ).

[0024] S16, using the rotating caliper algorithm to obtain the circumscribed rectangle vector information of the field.

[0025] In step S2, based on the field boundary information, a multi-task learning algorithm is used to extract field obstacles (such as trees and electric poles), and terrain features (such as slope) are obtained in combination with a digital elevation model to achieve a comprehensive analysis of the field attributes. In one embodiment, step S2 includes steps S21-S26.

[0026] S21, labeling dataset, collects remote sensing image datasets containing field obstacles, and performs precise labeling to generate labeling files.

[0027] S22 builds a multi-task model, using ResNet as the backbone network to extract shared underlying features of the image. Specific branches are set for different types of field obstacles, such as points, lines, and surfaces.

[0028] S23, feature fusion and multi-task model learning, uses a backbone network to extract multi-scale and multi-level features, which are then integrated through a feature fusion network. An attention mechanism is then used to weight the integrated features, enhancing the weights of key features and improving overall detection performance.

[0029] S24, obstacle detection, applies the above multi-task model to the remote sensing image of the target area, processes different types of obstacles in parallel, obtains obstacle information within the fields of the target area, and stores it in Shapefile format.

[0030] S25, use GIS tools combined with digital elevation models to obtain the topographic characteristics of the field (such as slope and elevation).

[0031] S26, use GIS tools to spatially overlay the obstacle detection results and terrain feature data with the field vectors.

[0032] In step S3, based on the driving characteristics of agricultural machinery in the field, a set of evaluation indicators that key factors affecting the driving efficiency of agricultural machinery are constructed, taking the field shape, obstacle distribution and terrain as the main factors, to ensure the targeted evaluation.

[0033] On the one hand, through demand analysis, combined with the characteristics of agricultural machinery operations and expert opinions, field morphology, obstacle distribution, and terrain characteristics were identified as core dimensions. On the other hand, during the indicator screening phase, candidate indicators were selected that were directly relevant to efficiency and quantifiable. Correlation analysis was performed to eliminate redundancies and ensure indicator independence. Based on these principles, field area, rectangularity, and morphological index were selected for field morphology; obstacle distribution within the field included the following: area percentage, number of obstacles, crowding, and minimum distance to the field boundary; and terrain included field slope and elevation variability. These nine indicators collectively constitute the key evaluation index set that influences agricultural machinery operation efficiency. Dimensionless values ​​for each indicator were calculated based on the field, obstacle, and terrain attribute information in S1 and S2.

[0034] The obtained indicator set not only reflects the geometric characteristics of the field, but also takes into account the impact of internal obstacles and terrain on the driving of agricultural machinery in the field. It can evaluate the suitability level of agricultural machinery driving efficiency at the field level in multiple dimensions, ensuring the comprehensiveness and objectivity of the evaluation process.

[0035] Table 1 Evaluation index set of key factors affecting the driving efficiency of agricultural machinery in the field

[0036]

[0037] In step S4, an indicator weighting method based on a geographically weighted attention mechanism is used. The spatial heterogeneity of farmland characteristics in various provinces in China is analyzed through a geographically weighted regression model to generate provincial-level benchmark weights. The attention mechanism is then used to dynamically adjust the weights based on the characteristics of specific plots, achieving dual-scale evaluation indicator weighting and significantly improving the regional adaptability of the weighting. In one embodiment, step S4 includes steps S41-S42.

[0038] S41 uses a geographically weighted regression model to analyze the spatial heterogeneity of farmland characteristics across China and generate provincial-level benchmark weights that reflect the differences in the importance of each indicator across provinces (e.g., large flat farmland plots in Heilongjiang versus small mountainous farmland plots in Guizhou).

[0039] Organize the data and obtain the farmland characteristic data of each province (i.e., the key evaluation indicators that affect the efficiency of agricultural machinery in the field). Let these data form a matrix X, where each row represents the farmland characteristic data of a province and each column corresponds to a farmland characteristic indicator. At the same time, obtain the historical data of agricultural machinery driving efficiency as a reference and set it as vector y. The elements of vector y correspond to the rows of matrix X, that is, the agricultural machinery driving efficiency y of the i-th provincei Corresponding to its farmland characteristic data X i Assuming there are n provinces and m farmland characteristic indicators, X is an n×m matrix and y is an n×1 vector.

[0040] (1) Construct a geographically weighted regression (GWR) model, using the agricultural machinery driving efficiency as the response variable and the farmland characteristic index set as the independent variable. For the i-th province, the model form is:

[0041]

[0042] (2) Among them, (u i ,v i ) is the geographical coordinate of the i-th province; β0(u i ,v i ) is the intercept term; β ik (u i ,v i ) is the regression coefficient of the kth independent variable (i.e., the kth farmland characteristic index) in the i-th province; x ik is the kth farmland characteristic index value of the i-th province; ∈ i is the error term. By fitting this model, we can obtain the regression coefficient β for each farmland characteristic index in each province. ik , the regression coefficients of all provinces form a coefficient matrix β, which is n×m in size.

[0043] (3) Calculate the provincial benchmark weights, normalize the regression coefficients of each province, and generate the provincial benchmark weights. For the kth indicator of the i-th province, its benchmark weight is The calculation formula is:

[0044]

[0045] Among them, |β ik | represents the absolute value of the regression coefficient, and the denominator represents the sum of the absolute values ​​of the regression coefficients of all indicators in the i-th province. It reflects the importance of the kth indicator relative to other indicators in the i-th province, and realizes the generation of benchmark weights that takes into account the spatial heterogeneity of provincial farmland characteristics.

[0046] Through the above steps, the geographically weighted regression model was used to analyze the spatial heterogeneity of farmland characteristics in various provinces in China, and provincial-level benchmark weights were generated to reflect the differences in the importance of each indicator in different provinces.

[0047] S42, based on the attention mechanism, calculates attention scores based on the characteristic values ​​of specific fields and obstacles (such as area size and slope steepness), dynamically adjusts weights, and ultimately integrates them into the final weights of all field-level indicators. This dual-scale weighting method helps improve the regional adaptability of weights, ensuring that the evaluation results are consistent with both the overall characteristics of the province and the individual characteristics of the fields. For example, in the Northeast Plain, the weight of field area is relatively high, while in the southern hills, the weights of slope and obstacle distribution are more prominent.

[0048] (1) Data preparation: collect characteristic data of specific fields in the province to form the characteristic value vector x t , with dimensions of 1×m.

[0049] (2) Index standardization: The Z-score standardization method is used to standardize the characteristic index values ​​of the fields.

[0050]

[0051] Among them, μ i is the mean value of the i-th indicator of all plots in the province, σ i is the standard deviation of the ith index of all plots in the province. The eigenvalue vector of the standardized characteristic index value of the plot is

[0052] (3) Calculate the attention score, use the attention mechanism to calculate the attention score of each indicator, and perform weight fusion. For the i-th indicator, its attention score a it See calculation formula 5 to get the attention score vector A t , the dimension is 1×m. The fusion formula is Formula 6.

[0053]

[0054]

[0055] Among them, α is the fusion coefficient, which ranges from [0,1]. The optimal value needs to be determined through sensitivity analysis and other methods. Finally, the final weight vector W at the field level is obtained. t , with dimensions of 1×m.

[0056] In step S5, based on the weighted indicators, a weighted summation method is used to calculate the suitability score of the agricultural machinery driving efficiency within the field, and a contribution analysis is performed to identify key negative impact indicators, providing content reference for specific field improvement matters. In one embodiment, step S5 includes steps S51-S54.

[0057] S51, data preparation, obtain the field-level weight w from step S4 tj , the standardized index value of the field Check the data to ensure that it is consistent and accurate in terms of format, statistical scope, etc., providing a reliable basis for subsequent calculations.

[0058] S52: Calculate the suitability score of the agricultural machinery driving efficiency in the field and perform statistical analysis. Use a weighted summation method to calculate the suitability score of the field and perform statistical analysis on the score results.

[0059] (1) Calculate the suitability score. For a certain field in a province, based on the field-level weight and standardized index value obtained in step S4, the weighted summation method is used to calculate the suitability score S of its agricultural machinery driving efficiency. t , the formula is as follows.

[0060]

[0061] Here w tj is the weight of the t-th field index j in the province obtained after fusion in step S4, is the value of the t-th field index j in the province after standardization. Through this formula, the corresponding suitability score of the fields in the province is calculated to obtain the score vector S t . Then, based on this method, the scoring vector of each field in each province is calculated.

[0062] (2) Statistical analysis: Statistical analysis is performed on the calculated suitability score results. Calculate the mean of the score It is used to reflect the average level of suitability of agricultural machinery driving efficiency of all fields in the province. The formula is as follows.

[0063]

[0064] Calculate the standard deviation σ S , measures the degree of dispersion of the score, and the formula is as follows.

[0065]

[0066] S53, locates key negative impact indicators, mainly by comparing the contribution values ​​of various indicators to screen out the negative factors that have the greatest impact on efficiency.

[0067] (1) Calculate the contribution of indicators. For the tth plot in a province, calculate the contribution of indicator j, C tj , the formula is as follows.

[0068]

[0069] in, is the optimal value of indicator j among all plots in the province (this needs to be determined based on the actual nature of the indicator; for example, for slope, a smaller value is optimal; for plot area, a larger value is optimal, etc.). This formula is used to calculate the contribution of each indicator for each plot in the province, yielding the contribution matrix C.

[0070] (2) Screen key negative impact indicators and compare the contribution C of each indicator of the tth plot in the province tj Size, find the indicator with the largest contribution, recorded as Then, the indicators with the highest frequency in each field in the province were counted and identified as the key negative factors that have the greatest impact on the efficiency of agricultural machinery in that province.

[0071] S54 maps the identified key negative indicators to the locations of specific farm plots within each province, taking into account the spatial distribution characteristics of farm plots within the province. Using geographic information system technology, the locations of key negative indicators are marked on a farm plot map of each province using different colors and symbols, creating a visualization. This intuitive display of the spatial distribution of key negative indicators across farm plots within a province provides a clear and intuitive reference for the subsequent development of targeted agricultural machinery operation optimization strategies.

[0072] In one embodiment, in step S6, the suitability score is divided into four levels: excellent, good, medium, and poor using the natural breakpoint method, providing scientific references such as location and quantification for specific field transformation.

[0073] The results calculated by the above evaluation model are between 0 and 1. According to the calculated scores, the evaluation results can be segmented using the natural breakpoint method.

[0074] S61, data preparation, obtain the agricultural machinery driving efficiency suitability score vector S of each field in the province calculated in step S5 t , the suitability score of the tth plot in the province.

[0075] S62, natural breakpoint method is used to divide the data into different levels. The natural breakpoint method is to cluster and group the data by traversing, so as to maximize the variance between groups and minimize the variance within groups, while considering that the range of each group and the elements within the group are as close as possible. First, obtain the average value of the suitability score of the plots in the province. and variance SDAM; again, iterate through the range combinations, calculate and find the smallest sum of squares of the category means SDCM; finally, calculate and judge the variance goodness of fit GVF. The value range of variance goodness of fit is 0-1, 1 means excellent fit, 0 means extremely poor fit, and the relevant calculation formula is as follows.

[0076]

[0077] Where t is the number of plots in the province; Si is the suitability score of the i-th field; k is the number of field category traversal groups in the province; t j represents the number of plots in group j; S ij Score the suitability of the i-th field in the j-th category; is the average suitability score of type j fields.

[0078] The natural breakpoint method described above was used to group similar values, and natural dividing points were set at the locations with the greatest difference in values. Taking the experimental plot as an example, the natural breakpoint method divided the agricultural machinery driving efficiency suitability scores within the field into four intervals, that is, k = 4, and the scores from low to high correspond to the four grades of poor, medium, good, and excellent.

[0079] S63, perform field classification mapping, use ArcGIS to add the field classification results to the field vector layer attribute table, and use the unique value method in the symbol system to visualize the field classification (see Figure 6 Based on the grading results, targeted agricultural machinery operation optimization strategies are developed for different farmland grades. For farmland with excellent grades, efficient agricultural machinery operation models can be further promoted to improve agricultural production efficiency. For farmland with poor grades, key negative impact indicators need to be analyzed and improvement measures implemented to enhance the suitability of agricultural machinery operation efficiency.

[0080] The present invention also provides an evaluation system for grading the suitability of agricultural machinery driving efficiency within a field, which includes a computer executable program and can complete the method steps described above.

[0081] Using the field and obstacle vector data obtained above, we can extract a variety of evaluation indicators related to the suitability of agricultural machinery driving efficiency within the field. These indicators are selected from the perspective of field morphological suitability: field area, rectangularity, and morphological index; from the perspective of field obstacle distribution suitability: obstacle area ratio, number, crowding, and minimum distance from the field boundary; and from the perspective of field terrain suitability: field slope and field elevation variability. These indicators are used as important indicators for evaluating the suitability of agricultural machinery driving efficiency within the field (Table 1).

[0082] The present invention meets the need for fine evaluation of the driving efficiency of agricultural machinery within a field, and proposes better evaluation indicators and weighting methods. The present invention uses the vector information of fields and field obstacles and terrain information in combination with the needs of agricultural machinery operations, and selects important evaluation indicators from three aspects: field shape, field obstacles and terrain. An indicator weighting method based on the geographically weighted attention mechanism is adopted to obtain the weight of each indicator. Finally, a suitable evaluation model for the driving efficiency of agricultural machinery within a field is constructed, and the fields are graded based on this. This innovative method can be widely used in modern agricultural machinery operations, promote the development of agricultural production methods towards intelligence and precision, and further improve agricultural production efficiency and sustainable development levels.

[0083] The embodiments described above are merely preferred embodiments of the present invention. The phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments" used in this specification may refer to one or more of the same or different embodiments of the present disclosure. Any common changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention are intended to be encompassed within the scope of protection of the present invention.

Claims

1. A method for evaluating the suitability of agricultural machinery driving efficiency within a field based on remote sensing images, comprising: S1, extracting detailed field attribute information based on high-resolution remote sensing images; S2, extracting field obstacle information and field terrain information based on field attribute information; S3, construct a set of key evaluation indicators that affect the driving efficiency of agricultural machinery in the field; S4, indicator weighting based on geographically weighted attention mechanism; S5, establish a suitability evaluation model for agricultural machinery driving efficiency within the field and identify key negative impact indicators; S6, Classification of suitability levels of agricultural machinery driving efficiency based on the natural breakpoint method.

2. The method according to claim 1, characterized in that Step S1 includes: S11, performing overlapping screening on the masks generated after segmentation of high-resolution remote sensing images; S12, performing a dilation and corrosion operation on the mask after the first segmentation, removing the linear non-cultivated land objects from the original image, and using the remaining unsegmented image area as input for the next segmentation; S13, performing iterative segmentation, performing initial instance segmentation on the high-resolution image using a rule-prompted point network with a specified density to generate an initial valid mask; S14, performing mask stitching of the predicted intersection-over-union ratio, and vectorizing the field raster mask result obtained by segmentation; S15, visualization of the final vector result of the field; S16, using the rotating caliper algorithm to obtain the circumscribed rectangle vector information of the field.

3. The method according to claim 1, characterized in that Step S14 includes: 1) Vectorize the segmented field raster mask results to separate and independently number each field mask; 2) The predicted IoU information generated during the segmentation process is transferred to the attributes of the vector elements, and all elements are sorted according to this information; 3) Traverse all elements, check the topological relationship in turn and add them to the result list. If the current element has a topological relationship with an existing element in the list, remove it to form the field area vector segmentation result.

4. The method according to claim 1, wherein Step S2 includes: S21, labeled dataset: collect and label remote sensing image datasets containing field obstacles; S22, build a multi-task model: use ResNet as the backbone network to extract the shared underlying features of the image; S23, Feature Fusion and Multi-task Model Learning: Use the backbone network to extract multi-scale and multi-level features, and fuse them through the feature fusion network; S24, obstacle detection: applying the multi-task model to the remote sensing image of the target area, processing different types of obstacles in parallel, and obtaining obstacle information within the field of the target area; S25, using GIS tools combined with digital elevation models to obtain the topographic characteristics of the fields; S26, use GIS tools to spatially overlay the obstacle detection results and terrain feature data with the field vectors.

5. The method according to claim 1, characterized in that In step S3, the evaluation index set includes: Evaluation indicators of field shape include: field area, rectangularity and shape index; Evaluation indicators of obstacle distribution include: area proportion, number, crowding degree and shortest distance from the field boundary; Evaluation indicators of terrain include: field slope and field elevation variability.

6. The method according to claim 1, characterized in that In step S4, an indicator weighting method based on the geographically weighted attention mechanism is adopted. The spatial heterogeneity of farmland characteristics in various provinces in China is analyzed through a geographically weighted regression model to generate provincial-level benchmark weights. The attention mechanism is then used to dynamically adjust the weights according to the specific characteristics of the fields to achieve dual-scale evaluation indicator weighting.

7. The method according to claim 6, characterized in that Step S4 includes: S41, using a geographically weighted regression model to analyze the spatial heterogeneity of farmland characteristics across China's provinces, generating provincial-level benchmark weights that reflect the differences in the importance of each indicator across provinces; S42, based on the attention mechanism, calculates the attention score according to the characteristic values ​​of specific fields and obstacles, dynamically adjusts the weight, and finally integrates it into the final weight of all indicators at the field level.

8. The method according to claim 7, characterized in that Step S5 includes: S51, data preparation, obtaining the field-level weights and the standardized index values ​​of the fields from step S4; S52, calculating the suitability score of the agricultural machinery driving efficiency in the field and conducting statistical analysis; S53, locate key negative impact indicators, mainly by comparing the contribution value of each indicator to screen out the negative factors that have the greatest impact on efficiency; S54, combined with the spatial distribution characteristics of farm plots within the province, maps the identified key negative indicators to the locations of specific farm plots in the province.

9. The method according to claim 8, characterized in that Step S6 includes: S61, data preparation, obtaining the agricultural machinery driving efficiency suitability score vector for each field in the province calculated in step S5; S62, using the natural breakpoint method for grading; S63, perform field grading mapping, add the field grading results to the field vector layer attribute table, and use the unique value method in the symbol system to visualize the field grading.

10. A field-level agricultural machinery driving efficiency suitability evaluation system based on remote sensing images, characterized in that: The method comprises computer executable codes, and when the codes are executed, the method according to any one of claims 1 to 9 is implemented.

Images