Machine learning-based intelligent matching and fertilization regulation planting method for farmland straw returning

Abstract

The present application relates to the field of intelligent agriculture and farmland fertilization regulation technology, and discloses a farmland straw returning intelligent proportioning fertilization regulation planting method based on machine learning; comprising: dividing fertilization subareas in the straw returning operation process, collecting straw returning amount, crushing length distribution, spreading uniformity, soil penetration depth, returning layer thickness, moisture content and carbon-nitrogen ratio, and establishing a returning layer state atlas; collecting soil stratified nutrient data, returning layer environment data and crop growth data, and inputting a straw decomposition state identification model to obtain the decomposition stage label of each fertilization subarea; determining the nitrogen competition window according to the decomposition stage label, the root zone mineral nitrogen content, the crop stage nitrogen requirement and the straw carbon-nitrogen ratio, and generating a subarea proportioning fertilization prescription; controlling the variable fertilization equipment to perform subarea fertilization according to the proportioning fertilization prescription, and updating the field sample library and the straw decomposition state identification model according to the feedback data after fertilization.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture and farmland fertilization regulation technology, specifically to a machine learning-based intelligent ratio fertilization regulation planting method for returning farmland straw to the field. Background Technology

[0002] Returning straw to the field is an important measure for improving farmland organic matter and utilizing agricultural waste resources. By crushing crop straw and returning it to the soil after harvest, soil organic carbon and some nutrients such as nitrogen, phosphorus, and potassium can be replenished, soil structure can be improved, and environmental problems caused by open burning of straw can be reduced. In current farmland planting management, nitrogen, phosphorus, and potassium application rates are usually determined after straw is returned to the field according to soil testing and formula fertilization, target yield fertilization, or empirical fertilization methods. These methods mainly rely on soil nutrient test values, crop nutrient requirements, and target yield levels to determine fertilizer ratios, and are applicable to conventional fields where straw is not returned to the field.

[0003] However, the nutrient conversion process after straw is returned to the field is not immediately synchronized with the crop's nutrient requirements. Differences in the amount of straw returned, the length of the crushed straw, the evenness of spreading, the depth of burial, the moisture content, the carbon-nitrogen ratio, and the temperature and humidity of the returned layer can lead to significant differences in the straw decomposition rate between different plots and even different areas within the same plot. Especially in the early stages of straw decomposition, microorganisms decompose straw with a high carbon-nitrogen ratio, fixing mineral nitrogen in the soil. This can easily compete with the available nitrogen for crop seedling growth, resulting in insufficient nitrogen supply in the root zone in the early stages. In the later stages of decomposition, nutrients such as nitrogen and potassium in the straw are gradually released. If conventional topdressing methods are still applied, it can easily lead to problems such as excessive topdressing in the later stages and a mismatch between nutrient supply and crop absorption.

[0004] While existing intelligent fertilization or variable fertilization technologies can collect data on soil nutrients, crop growth, and meteorological conditions, and generate fertilization recommendations or prescription maps accordingly, most schemes still rely primarily on crop nutrient requirements, soil nutrient deficits, or target yields as the main decision-making basis. They fail to consider the state of the straw return layer after straw return, changes in the decomposition stage, the enhanced period of microbial nitrogen fixation, and the release process of straw-derived nutrients as core intermediate variables for fertilization regulation. Consequently, under straw return conditions, existing methods struggle to accurately identify nitrogen competition windows in different fertilization zones and to dynamically adjust the nitrogen allocation of decomposition-initiating compensating fertilizer, seedling topdressing, and subsequent topdressing based on the straw decomposition stage, resulting in a lack of effective synergy between fertilization prescriptions and the straw decomposition process. Therefore, it is necessary to propose a crop cultivation method that integrates straw return status, decomposition stage identification, nitrogen competition window judgment, and variable fertilization execution feedback for straw ratio fertilization regulation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent method for regulating and controlling crop planting by returning farmland straw to the field based on machine learning, in order to solve the technical problems existing in the prior art.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution: A machine learning-based intelligent method for regulating crop growth through straw return and fertilization includes the following steps: S1. During the straw return operation, the target farmland is divided into multiple fertilization zones. Straw return operation data of each fertilization zone is collected, and a return layer state map of the corresponding fertilization zone is established based on the straw return operation data. The straw return operation data includes the amount of straw returned, the distribution of straw crushing length, the uniformity of straw spreading, the depth of straw entering the soil, the thickness of the return layer, the straw moisture content, and the straw carbon-nitrogen ratio. S2. Before sowing and at the pre-set growth stage of the crop, collect soil stratified nutrient data, return layer environmental data and crop growth data for each fertilization zone. Soil stratified nutrient data includes mineral nitrogen content, available phosphorus content, available potassium content and organic matter content. Return layer environmental data includes return layer temperature, return layer moisture content and return layer redox potential. S3. Input the straw decomposition status map, soil stratification nutrient data, straw decomposition environment data and crop growth data into the straw decomposition status recognition model trained by field samples to obtain the straw decomposition stage label corresponding to each fertilization zone. The straw decomposition stage label includes the decomposition initiation period, the microbial nitrogen fixation enhancement period, the slow nitrogen release period and the nutrient release compensation period. S4. Based on the straw decomposition stage label, root zone mineral nitrogen content, crop stage nitrogen requirement and straw carbon-nitrogen ratio, determine the nitrogen competition window for each fertilization zone, and determine the nitrogen fertilizer allocation for each fertilization zone in basal fertilizer, decomposition start-up compensation fertilizer, seedling topdressing, rapid vegetative growth period topdressing and yield formation early topdressing. Decomposition start-up compensation fertilizer is the nitrogen fertilizer unit applied from straw return to the field to before the seedling stage. S5. Based on the predicted release of nitrogen from straw sources, the predicted release of potassium from straw sources, the deficit of available nutrients in the soil, and the fertilizer requirements of crops at each fertilization zone, generate the corresponding fertilizer formula for the fertilization zone. The fertilizer formula includes the nitrogen-phosphorus-potassium ratio, fertilizer application time, fertilizer form, fertilization depth, and fertilizer application rate per unit area. S6. Apply fertilizer to the target farmland in zones according to the fertilizer formula and control variable fertilization equipment. After fertilization, collect data on the change of mineral nitrogen in the root zone, crop growth feedback data and yield data at harvest time for each fertilization zone. Update the field sample library based on the collected results and retrain the straw decomposition status recognition model using the updated field sample library.

[0007] Preferably, in S1, when dividing the target farmland into multiple fertilization zones, the boundary coordinates and operation path coordinates of the target farmland are first obtained, and then the coordinate range of the fertilization zone is determined according to the operation width of the straw returning machine, the single fertilizer discharge control width of the variable fertilizer equipment, and the operation path direction, and each fertilization zone is assigned a zone number. The straw return layer status map includes the fertilization zone number, the coordinate range of the fertilization zone, and the straw return operation data corresponding to the fertilization zone.

[0008] Preferably, in S1, the amount of straw returned to the field is determined by harvester output monitoring data, crop straw-to-grain ratio, and straw-to-field ratio; The length distribution of the crushed straw was extracted from the surface image after the straw was returned to the field; The uniformity of straw spreading is determined by the ratio of the straw coverage area to the area of ​​the fertilization zone within the same fertilization zone; The straw penetration depth and the thickness of the returned layer are obtained by correcting the values ​​detected by the tillage depth sensor and the soil profile sampling values.

[0009] Preferably, in S2, the soil nutrient data are collected from the 010cm soil layer, the 1020cm soil layer, and the 2040cm soil layer, respectively. The preset growth period nodes of the crop include pre-sowing, seedling stage, rapid vegetative growth stage, pre-yield formation stage and harvest stage, and the preset growth period nodes of the crop are determined according to the cultivation technical regulations corresponding to the target crop. The crop growth data includes leaf color index, canopy coverage, leaf area index, and plant nitrogen nutrition index.

[0010] Preferably, in S3, the field samples include the return layer state map data, soil stratification nutrient data, return layer environmental data, crop growth data, and straw decomposition stage labels of the same fertilization zone at the same sampling node; The straw decomposition stage labels are based on the percentage of residual straw mass, changes in mineral nitrogen in the returned layer, changes in soluble organic carbon in the returned layer, and changes in the straw carbon-nitrogen ratio. Specifically, the decomposition initiation stage is marked when the percentage of residual straw mass begins to decrease and the soluble organic carbon in the returned layer increases. The microbial nitrogen fixation enhancement stage is marked when the mineral nitrogen content in the root zone continuously decreases and the straw carbon-nitrogen ratio remains no less than 0.70 times the straw carbon-nitrogen ratio before sowing. The slow nitrogen release stage is marked when the percentage of residual straw mass continues to decrease and the mineral nitrogen content in the root zone stops decreasing continuously. The nutrient release compensation stage is marked when the mineral nitrogen content in the root zone reaches the lower limit of nitrogen supply in the root zone and the straw carbon-nitrogen ratio is less than 0.70 times the straw carbon-nitrogen ratio before sowing.

[0011] Preferably, in S3, the straw decomposition state identification model is a classification model, which includes a feature preprocessing module, a classifier, and an output module. The feature preprocessing module standardizes or normalizes the input features, which include straw return amount, straw crushing length distribution, straw penetration depth, return layer temperature, return layer moisture content, return layer redox potential, root zone mineral nitrogen content, available potassium content, leaf color index, and canopy coverage. The classifier uses any one of the following: random forest classifier, gradient boosting tree classifier, or support vector machine classifier. The output module outputs a label for one of the straw decomposition stages: decomposition initiation period, microbial nitrogen fixation enhancement period, slow nitrogen release period, and nutrient release compensation period.

[0012] Preferably, in S4, the nitrogen requirement of the crop stage is determined based on the target crop type, target yield level, and preset growth period of the crop; The lower limit of nitrogen supply in the root zone is determined based on the nitrogen requirement of the crop stage, the thickness of the soil layer for effective root absorption, soil bulk density, and the area of ​​the fertilization zone. The mineral nitrogen content in the root zone is obtained by adding the nitrate nitrogen content and ammonium nitrogen content in the soil layer where the roots can effectively absorb nitrogen.

[0013] Preferably, in S4, the starting node of the nitrogen competition window is the earlier sampling node when two consecutive sampling nodes in the same fertilization zone meet the following conditions: the straw decomposition stage is labeled as the microbial nitrogen fixation enhancement period, the mineral nitrogen content in the root zone is lower than the lower limit of nitrogen supply in the root zone, and the straw carbon-nitrogen ratio is not lower than 0.70 times the straw carbon-nitrogen ratio before sowing in the fertilization zone. The termination node of the nitrogen competition window is the sampling node when the mineral nitrogen content in the root zone of the fertilization zone reaches the lower limit of nitrogen supply in the root zone, and the straw decomposition stage label changes to the slow nitrogen release period or nutrient release compensation period.

[0014] Preferably, in S4, when the fertilization zone is within the nitrogen competition window, the difference between the mineral nitrogen content in the root zone and the lower limit of nitrogen supply in the root zone is included in the nitrogen fertilizer allocation of the decomposition start-up compensation fertilizer and the seedling topdressing. When the fertilization zone is in the nutrient release compensation period, the amount of nitrogen fertilizer allocated for topdressing during the rapid vegetative growth period or the early yield formation period is reduced according to the predicted amount of nitrogen release from straw sources.

[0015] Preferably, in S5, the predicted amount of nitrogen release from straw source and the predicted amount of potassium release from straw source are determined based on the amount of straw dry matter returned to the field, the straw nitrogen content, the straw potassium content, and the nutrient release coefficient corresponding to the straw decomposition stage label of the corresponding fertilization zone, respectively. The straw nitrogen content and straw potassium content are obtained by detecting straw samples before returning to the field. The nutrient release coefficient is determined based on the percentage of residual straw mass, the change in straw nitrogen content, and the change in straw potassium content corresponding to the same straw decomposition stage label in the field sample bank. In S6, the variable fertilization equipment generates fertilization control instructions according to the fertilization zone number, fertilizer amount per unit area, fertilizer form and fertilization depth in the fertilization formula, and adjusts the speed of the fertilizer discharge motor, the opening of the fertilizer discharge port and the soil penetration depth of the furrow opener according to the fertilization control instructions.

[0016] In summary, the present invention has the following main beneficial effects: By collecting data on the amount of straw returned to the field, the distribution of crushed length, the uniformity of spreading, the depth of penetration into the soil, the thickness of the returned layer, the moisture content, and the carbon-nitrogen ratio during the straw returning operation, and linking the above data with soil stratification nutrient data, returned layer environmental data, and crop growth data, a returned layer state map corresponding to the fertilization zone is formed. This achieves the effect of regional characterizing the spatial differences after straw returning, the basic conditions of decomposition, and the nutrient requirements of crops. This avoids the fertilization deviation caused by the uneven straw returning, different penetration depths, and different decomposition rates caused by traditional soil testing and formula fertilization, which only determines the total amount of fertilizer based on the average nutrient content or target yield of the plot.

[0017] By utilizing a straw decomposition state identification model, labels for straw decomposition stages such as decomposition initiation, microbial nitrogen fixation enhancement, slow nitrogen release, and nutrient release compensation are obtained. Based on these decomposition stage labels, root zone mineral nitrogen content, crop nitrogen requirements at different stages, and the straw carbon-nitrogen ratio, a nitrogen competition window is determined. This transforms the insufficient nitrogen supply in the root zone caused by enhanced microbial nitrogen fixation in the early stages of straw return into an identifiable, predictable, and actionable basis for fertilization. Consequently, the amount of nitrogen fertilizer allocated for decomposition initiation compensation fertilizer and seedling topdressing can be rationally determined within the nitrogen competition window. Furthermore, during the nutrient release compensation period, the amount of subsequent topdressing is reduced based on the predicted amount of nitrogen release from straw sources, thus mitigating the coexistence of early-stage nitrogen deficiency and late-stage excessive topdressing.

[0018] By generating zoned fertilization prescriptions based on predicted nitrogen and potassium release from straw, soil available nutrient deficit, and crop stage nutrient requirements, variable fertilization equipment executes fertilization according to fertilization zone number, fertilizer application rate per unit area, fertilizer form, and fertilization depth. Simultaneously, actual fertilizer discharge, root zone mineral nitrogen change data, crop growth feedback data, and harvest yield data are written into a field sample database. This achieves a closed-loop effect of fertilization prescription generation, variable fertilization execution, and model feedback updates. In subsequent planting seasons, the straw decomposition status identification model and nutrient release coefficient can be corrected based on the actual decomposition process and fertilization feedback of similar plots, improving the adaptability of fertilization regulation under straw return conditions and the stability of continuous planting management. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] refer to Figure 1 A machine learning-based intelligent method for regulating crop planting by combining fertilizer application with straw return to the field includes the following steps: S1. During the straw return operation, the target farmland is divided into multiple fertilization zones. Straw return operation data of each fertilization zone is collected, and a return layer state map of the corresponding fertilization zone is established based on the straw return operation data. The straw return operation data includes the amount of straw returned, the distribution of straw crushing length, the uniformity of straw spreading, the depth of straw entering the soil, the thickness of the return layer, the straw moisture content, and the straw carbon-nitrogen ratio. S2. Before sowing and at the pre-set growth stage of the crop, collect soil stratified nutrient data, return layer environmental data and crop growth data for each fertilization zone. Soil stratified nutrient data includes mineral nitrogen content, available phosphorus content, available potassium content and organic matter content. Return layer environmental data includes return layer temperature, return layer moisture content and return layer redox potential. S3. Input the straw decomposition status map, soil stratification nutrient data, straw decomposition environment data and crop growth data into the straw decomposition status recognition model trained by field samples to obtain the straw decomposition stage label corresponding to each fertilization zone. The straw decomposition stage label includes the decomposition initiation period, the microbial nitrogen fixation enhancement period, the slow nitrogen release period and the nutrient release compensation period. S4. Based on the straw decomposition stage label, root zone mineral nitrogen content, crop stage nitrogen requirement and straw carbon-nitrogen ratio, determine the nitrogen competition window for each fertilization zone, and determine the nitrogen fertilizer allocation for each fertilization zone in basal fertilizer, decomposition start-up compensation fertilizer, seedling topdressing, rapid vegetative growth period topdressing and yield formation early topdressing. Decomposition start-up compensation fertilizer is the nitrogen fertilizer unit applied from straw return to the field to before the seedling stage. S5. Based on the predicted release of nitrogen from straw sources, the predicted release of potassium from straw sources, the deficit of available nutrients in the soil, and the fertilizer requirements of crops at each fertilization zone, generate the corresponding fertilizer formula for the fertilization zone. The fertilizer formula includes the nitrogen-phosphorus-potassium ratio, fertilizer application time, fertilizer form, fertilization depth, and fertilizer application rate per unit area. S6. Apply fertilizer to the target farmland in zones according to the fertilizer formula and control variable fertilization equipment. After fertilization, collect data on the change of mineral nitrogen in the root zone, crop growth feedback data and yield data at harvest time for each fertilization zone. Update the field sample library based on the collected results and retrain the straw decomposition status recognition model using the updated field sample library.

[0022] The method described in this embodiment is applicable to the regulation of crop planting in farmland after straw return, where the farmland can be planted with crops such as wheat, corn, rice, and rapeseed. The specific planting stages for different crops—before sowing, seedling stage, rapid vegetative growth stage, early yield formation stage, and harvest period—are determined according to the cultivation technical regulations, soil testing and fertilizer recommendation data, agricultural extension data, or historical planting records of the target farmland. The straw can be corn straw, wheat straw, rice straw, or rapeseed straw.

[0023] In this embodiment, fertilization regulation does not involve directly inputting soil nutrient, meteorological, and crop growth data into a general fertilization model to output the fertilization amount. Instead, it first establishes a state map of the straw return layer based on straw return operation data, then identifies straw decomposition stage labels, and determines the nitrogen competition window based on straw decomposition stage labels, root zone mineral nitrogen status, and crop nitrogen requirements at each stage. Finally, it generates fertilization prescriptions with zoned, staged, and form-specific ratios. This process ensures that the straw decomposition process after return to the field, the enhanced microbial nitrogen fixation process, and the subsequent release of straw-derived nutrients all participate in fertilization regulation, avoiding the limitation of determining the total fertilization amount solely based on target yield or soil nutrient surplus / deficit.

[0024] Before straw return to the field, the boundary coordinates of the target farmland, the coordinates of the operation path, the operating width of the straw return machinery, and the single-time fertilizer application control width of the variable fertilizer application equipment are obtained. The boundary coordinates of the target farmland can be obtained from a Beidou positioning terminal, an agricultural machinery autopilot system, UAV aerial survey boundaries, or an existing electronic map of the land. Based on the boundary coordinates of the target farmland, the direction of the operation path, the operating width of the straw return machinery, and the single-time fertilizer application control width of the variable fertilizer application equipment, the target farmland is divided into multiple fertilization zones, and each fertilization zone is assigned a zone number.

[0025] Each fertilization zone has a defined coordinate range. The width of the fertilization zone corresponds to the fertilizer discharge width that the variable fertilization equipment can independently control, and the length of the fertilization zone is divided along the direction of the agricultural machinery's operating path. Through this division method, subsequent fertilizer formulations can be directly mapped to the fertilizer discharge control unit of the variable fertilization equipment, avoiding inconsistencies between the formulation zones and the equipment's execution zones.

[0026] During the straw return to the field operation, straw return operation data for each fertilization zone is collected. The straw return operation data includes the amount of straw returned, the distribution of straw crushing length, the uniformity of straw spreading, the depth of straw penetration into the soil, the thickness of the returned layer, the straw moisture content, and the straw carbon-nitrogen ratio.

[0027] The amount of straw returned to the field can be determined by harvester yield monitoring data, the crop straw-to-grain ratio, and the straw return ratio. When the harvester has yield monitoring capabilities, the grain yield monitoring data corresponding to the fertilization zone is used as the basic data; when the harvester does not have yield monitoring capabilities, it can be determined through measured yield in sample plots, agricultural machinery operation weighing data, or historical yield records of the same plot. The crop straw-to-grain ratio and the straw return ratio are determined using local agricultural technology data, operation records, or field sample plot testing data. When the fresh straw mass is obtained on-site, it is first converted into straw dry matter mass according to the straw moisture content before being used as the amount of straw returned to the field.

[0028] The distribution of straw crushing length was extracted from surface images after straw was returned to the field or from on-site sampling images. Specifically, multiple sampling points were selected within each fertilization zone to collect straw-covered images including a scale. Background segmentation, straw region identification, and connected component extraction were performed on the images to obtain the pixel length of the straw segments in the image, which was then converted to the actual length using the image scale. For straw segments with severe shading or obvious stacking, corrections were made through manual verification or quadrat sampling. The straw crushing length distribution includes at least the mean straw segment length, the median length, and the proportion of different length intervals.

[0029] The uniformity of straw spreading is determined by the ratio of the straw coverage area to the area of ​​the same fertilization zone. The straw coverage area is calculated from the area of ​​straw regions identified in the surface image. To reduce the influence of shadows, local accumulation, or shooting angle on the judgment results, the weighted average of the straw coverage area ratios of multiple sampling points within the same fertilization zone is used as the straw spreading uniformity of that fertilization zone.

[0030] The depth of straw penetration and the thickness of the returned layer are obtained by correcting the values ​​detected by the tillage depth sensor and the values ​​obtained by soil profile sampling. The tillage depth sensor is installed on the straw returning machine or land preparation machine to record the tillage depth during the operation. After the operation, a soil profile is excavated in the fertilization zone to observe the upper and lower boundaries of the concentrated straw distribution layer, thus obtaining the straw penetration depth and the thickness of the returned layer. If there is a deviation between the tillage depth sensor value and the profile sampling value, the profile sampling value is used as the correction basis to correct the sensor value under similar operating conditions.

[0031] Straw moisture content, straw carbon-to-nitrogen ratio, straw nitrogen content, and straw potassium content were obtained from straw samples collected before or shortly after returning straw to the field. Straw samples were taken from mixed samples according to fertilization zones or the same batch. Straw moisture content could be measured by drying or using calibrated rapid moisture detection equipment; straw carbon, nitrogen, and potassium content could be measured by laboratory testing or using calibrated agricultural rapid testing equipment. The straw carbon-to-nitrogen ratio was determined by the ratio of straw carbon content to straw nitrogen content.

[0032] After completing the above data collection, the zone number, coordinate range, straw return amount, straw crushing length distribution, straw spreading uniformity, straw penetration depth, return layer thickness, straw moisture content, and straw carbon-nitrogen ratio of each fertilization zone are associated and stored to form a return layer state map. The return layer state map uses the fertilization zone as the basic unit and is used for subsequent identification of straw decomposition stages, determination of nitrogen competition windows, and generation of fertilization prescriptions.

[0033] Before sowing and at the pre-set growth stages of the crop, soil nutrient data, environmental data of the returned soil layer, and crop growth data are collected for each fertilization zone. The pre-set growth stages of the crop include pre-sowing, seedling stage, rapid vegetative growth stage, early yield formation stage, and harvest stage. The specific stages are determined according to the cultivation technical regulations corresponding to the target crop.

[0034] Soil stratified nutrient data were collected from the 0-10cm, 10-20cm, and 20-40cm soil layers. Multiple sampling points were selected within each fertilization zone using either an S-shaped or grid sampling method. Soil samples from the same layer were mixed and then analyzed to obtain the mineral nitrogen, available phosphorus, available potassium, and organic matter content. Mineral nitrogen content includes nitrate nitrogen and ammonium nitrogen. Detection methods can employ commonly used chemical analysis methods in agricultural soil testing, or rapid field testing equipment calibrated based on laboratory test results.

[0035] Environmental data for the straw return layer includes temperature, moisture content, and redox potential. Temperature and moisture content can be detected by temperature and humidity sensors buried within the straw return layer, or by manual sampling combined with a soil moisture meter. The redox potential is measured using a redox potential electrode. Sensor placement corresponds to the layer where straw is concentrated, avoiding the use of surface air temperature or deep, non-returned soil conditions as environmental data for the straw return layer.

[0036] Crop growth data includes leaf color index, canopy coverage, leaf area index, and plant nitrogen nutrient index. Leaf color index can be obtained from handheld chlorophyll meters, canopy images, or multispectral images; canopy coverage can be obtained from UAV images or ground image segmentation; leaf area index can be obtained from canopy analyzers, remote sensing inversion results, or quadrat measured data; plant nitrogen nutrient index can be determined based on crop dry matter accumulation and plant nitrogen concentration, or it can be obtained by converting nitrogen nutrient diagnostic models established in local agricultural technology trials.

[0037] After the above data collection is completed, it is stored according to the fertilization zone number, sampling node and coordinate range to form the data foundation for model training, decomposition stage judgment and fertilization prescription generation.

[0038] Data from the same fertilization zone at the same sampling node constitutes one field sample. Each field sample includes data on the state of the straw return layer, soil stratified nutrient data, straw return layer environmental data, crop growth data, and straw decomposition stage labels. The straw decomposition stage labels include the decomposition initiation period, the microbial nitrogen fixation enhancement period, the slow nitrogen release period, and the nutrient release compensation period.

[0039] The labeling for the straw decomposition stage includes the percentage of straw residue mass, the change in mineral nitrogen in the returned layer, the change in soluble organic carbon in the returned layer, and the change in the straw carbon-nitrogen ratio. The percentage of straw residue mass can be measured using the mesh bag method, the fixed-plot residual straw recovery method, or the soil profile sampling and separation method. When using the mesh bag method, straw samples with known dry matter mass are placed in a water-permeable and air-permeable mesh bag and buried in the returned layer. The samples are then removed at sampling points, washed, dried, and weighed to calculate the percentage of straw residue mass. When using the fixed-plot method, a mixed soil and straw sample is collected from a fixed plot, and the dry matter mass is measured after separating the straw residue. Soluble organic carbon in the returned layer is obtained by extracting from the returned layer soil sample and then detecting it. The change in the straw carbon-nitrogen ratio is determined by the measured carbon and nitrogen content of the straw residue at different sampling points.

[0040] When labeling the straw decomposition stage, the decrease in the proportion of residual straw mass, the increase in soluble organic carbon in the returned layer, the decrease in mineral nitrogen content in the root zone, and the change in the straw carbon-nitrogen ratio are all based on the effective changes between adjacent sampling nodes. An effective change refers to a change in the detected value exceeding the allowable error of the detection method used. The allowable error of the detection method is determined by the calibration record of the detection equipment, the repeatability error of the laboratory detection method, or the results of parallel tests on the same sample. If the change does not exceed the allowable error of the detection method, the indicator is considered to have not undergone an effective change between adjacent sampling nodes. For the detection results of multiple sampling points within the same fertilization zone, a weighted average is taken according to the area represented by each sampling point, and this average is used as the judgment value for that fertilization zone.

[0041] The specific labeling rules are as follows: When the proportion of residual straw mass effectively decreases and the soluble organic carbon in the returned layer effectively increases, the field sample is marked as the decomposition initiation period.

[0042] When the mineral nitrogen content in the root zone decreases effectively at two consecutive sampling points, and the straw carbon-nitrogen ratio remains no less than 0.70 times the straw carbon-nitrogen ratio before sowing, the field sample is marked as the period of enhanced microbial nitrogen fixation.

[0043] When the proportion of residual straw mass continues to decrease effectively, and the mineral nitrogen content in the root zone no longer meets the condition of effective decrease for two consecutive sampling nodes, the field sample is marked as the slow nitrogen release period.

[0044] When the mineral nitrogen content in the root zone reaches the lower limit of nitrogen supply in the root zone, and the carbon-nitrogen ratio of straw is less than 0.70 times that of straw before sowing, the field sample is marked as the nutrient release compensation period.

[0045] The aforementioned 0.70 is a relative judgment value, based on the straw carbon-nitrogen ratio before sowing in the same fertilization zone, rather than an absolute carbon-nitrogen ratio threshold applicable to straw from different crops. This relative judgment value can be determined initially based on field decomposition test data, local agricultural technology data, or existing test records from the applicant, under similar straw types, return methods, and environmental conditions in the return layer. Once the field sample bank has accumulated data for a planting season, this relative judgment value is corrected based on the decomposition stage markings in the field sample bank. In actual judgment of the decomposition stage, 0.70 is not used as the sole criterion; instead, it is determined by simultaneously considering the percentage of residual straw mass, changes in mineral nitrogen content in the root zone, and changes in soluble organic carbon in the return layer.

[0046] The straw decomposition status identification model is a classification model, including a feature preprocessing module, a classifier, and an output module. The feature preprocessing module is used to standardize or normalize the input features. Input features include straw return to the field amount, straw crushing length distribution, straw penetration depth, temperature of the returned layer, moisture content of the returned layer, redox potential of the returned layer, mineral nitrogen content in the root zone, available potassium content, leaf color index, and canopy coverage.

[0047] The classifier employs any one of the following: random forest classifier, gradient boosting tree classifier, or support vector machine classifier. During model training, field samples with completed decomposition stage labeling are divided into training samples and validation samples. Training samples are used to establish the correspondence between input features and straw decomposition stage labels, while validation samples are used to check the consistency between the model output and the field labeling results. When the model's identification results are inconsistent with the field labeling results, priority is given to checking the sampled data, labeling, and outlier samples, rather than simply increasing model complexity to replace field data correction.

[0048] The output module outputs a label for one of the straw decomposition stages: the decomposition initiation stage, the microbial nitrogen fixation enhancement stage, the slow nitrogen release stage, and the nutrient release compensation stage. The output results are stored in correspondence with the fertilization zone number, sampling node, and model input data, serving as the basis for subsequent nitrogen competition window judgment and proportional fertilization prescription generation.

[0049] In this embodiment, the straw decomposition status identification model does not directly output the fertilizer application rate, but rather outputs a label for the straw decomposition stage. Determining the fertilizer application rate also requires considering the root zone mineral nitrogen content, the lower limit of nitrogen supply to the root zone, the crop's nitrogen requirement at each stage, the predicted nitrogen release from straw sources, and the predicted potassium release from straw sources. Therefore, the model identification results correspond to the farmland straw decomposition process and the implementation control of fertilization.

[0050] The mineral nitrogen content in the root zone is obtained by adding the nitrate nitrogen content and ammonium nitrogen content in the soil layer where the roots can effectively absorb nitrogen. The thickness of the soil layer where the roots can effectively absorb nitrogen is determined according to the target crop species and the preset growth stage of the crop, and its source is local cultivation technical regulations, agricultural technology test data, or actual measurement records of the target farmland. The thickness of the soil layer where the roots can effectively absorb nitrogen can vary at different growth stages.

[0051] When soil test results are expressed in mg / kg, the mineral nitrogen content in the root zone is converted to nitrogen content per unit area using the following formula: ; in, For the first The fertilization zone is in the first Mineral nitrogen content in the root region of each sampling node, in kg / hm² For the first The first fertilization zone The soil layer in the first The nitrate nitrogen content at each sampling point is expressed in mg / kg. For the first The first fertilization zone The soil layer in the first The ammonium nitrogen content at each sampling node, in mg / kg; For the first The first fertilization zone The soil bulk density of each soil layer, in g / cm³ For the first Each soil layer thickness is expressed in cm. This represents the number of soil layers included in the calculation within the effective root absorption soil layer; 0.1 is a unit conversion factor used to convert mg / kg to g / c. Convert cm to kg / hm2.

[0052] The root zone nitrogen supply limit is the minimum amount of mineral nitrogen that the roots should effectively absorb from the soil layer to maintain nitrogen supply in the root zone during the corresponding growth stage of a target crop. The root zone nitrogen supply limit is not a fixed empirical value; its sources include the target crop species, target yield level, local soil testing and fertilizer recommendation indicators, nitrogen requirement patterns of the target crop during the corresponding growth stage, historical yield records of the target farmland, and multi-year experimental data from similar plots.

[0053] When determining the specific requirements, first determine the nitrogen requirement for each crop stage based on the target crop type, target yield level, and pre-set growth period. Then, combine the effective root absorption soil layer thickness, soil bulk density, and fertilization zone area to determine the required mineral nitrogen level to be maintained in the root zone for the corresponding growth period as the lower limit of nitrogen supply to the root zone. When local soil testing and fertilizer recommendation data or multi-year experimental data of similar plots already contain root zone mineral nitrogen supply indicators for the corresponding growth period, these indicators should be used as the initial lower limit of nitrogen supply to the root zone, and corrected at harvest time based on crop growth feedback, yield data, and post-harvest soil residual mineral nitrogen data.

[0054] The nitrogen competition window refers to the time interval during which microbial fixation of mineral nitrogen increases during straw decomposition, while the supply of mineral nitrogen in the crop root zone is lower than the nitrogen requirement for the corresponding growth stage. This window is used to determine the nitrogen fertilizer allocation among decomposition initiation compensating fertilizer, seedling topdressing, rapid vegetative growth topdressing, and pre-yield formation topdressing.

[0055] When two consecutive sampling nodes in the same fertilization zone meet the following conditions, the earlier sampling node will be designated as the starting node of the nitrogen competition window: the straw decomposition stage is labeled as the microbial nitrogen fixation enhancement period; the mineral nitrogen content in the root zone is lower than the lower limit of nitrogen supply in the root zone; and the straw carbon-nitrogen ratio is not lower than 0.70 times the straw carbon-nitrogen ratio before sowing in that fertilization zone. Setting two consecutive sampling nodes as the judgment condition is to avoid misjudgment caused by single detection errors, local moisture fluctuations, or short-term fertilization disturbances.

[0056] When the mineral nitrogen content in the root zone of a fertilization zone reaches the lower limit of nitrogen supply from the root zone, and the straw decomposition stage label changes to the slow nitrogen release period or nutrient release compensation period, the sampling node is determined as the termination node of the nitrogen competition window. If the judgment results of multiple sampling points within the same fertilization zone are different, a weighted average is calculated according to the area represented by the sampling points, and the weighted result is used as the basis for judgment of the fertilization zone.

[0057] Therefore, the nitrogen competition window is not determined solely by the straw carbon-nitrogen ratio, but is jointly determined by the straw decomposition stage label, root zone mineral nitrogen content, root zone nitrogen supply limit, and the relative change in the straw carbon-nitrogen ratio.

[0058] Decomposition-initiating compensating fertilizer is a nitrogen fertilizer unit applied from straw return to the field until the seedling stage. Its application is based on the risk of a nitrogen competition window forming or an already formed nitrogen competition window, rather than being a fixed component of ordinary basal fertilizer. Decomposition-initiating compensating fertilizer differs from seedling topdressing; seedling topdressing is mainly determined based on crop growth and root zone nutrient monitoring results, while decomposition-initiating compensating fertilizer is used to compensate for the decrease in root zone nitrogen supply caused by microbial fixation of mineral nitrogen during the initial stage of straw decomposition.

[0059] When the fertilization zone is within the nitrogen competition window, the difference between the mineral nitrogen content in the root zone and the lower limit of nitrogen supply in the root zone is determined according to the following formula: ; in, For the first The fertilization zone is in the first Nitrogen supply difference in the root region of each sampling node, in kg / hm; For the first The fertilization zone is in the first The lower limit of nitrogen supply in the root region of each sampling node, in kg / hm; For the first The fertilization zone is in the first The mineral nitrogen content in the root region of each sampling node is expressed in kg / hm. This indicates taking the larger value between the number inside the parentheses and 0.

[0060] When the nitrogen competition window starts earlier than the seedling sampling window, the difference in nitrogen supply to the root zone is preferentially included in the decomposition initiation compensation fertilizer application. When the nitrogen competition window starts after the seedling sampling window, or the decomposition initiation compensation fertilizer application window has ended, the difference in nitrogen supply to the root zone is included in the seedling topdressing. If the nitrogen competition window spans both the decomposition initiation compensation fertilizer application window and the seedling topdressing window, the difference in nitrogen supply to the root zone is allocated according to the proportion of the number of consecutive days of the two fertilization windows within the nitrogen competition window.

[0061] When the fertilization zone is in the nutrient release compensation period, the amount of nitrogen fertilizer allocated in topdressing during the rapid vegetative growth period or the early yield formation period is deducted based on the predicted amount of nitrogen release from straw sources, the measured value of mineral nitrogen in the root zone, and the crop growth feedback data. If the mineral nitrogen content in the root zone is still lower than the lower limit of nitrogen supply in the root zone, the topdressing will not be directly deducted based on the predicted amount of nitrogen release from straw sources; if the mineral nitrogen content in the root zone reaches the lower limit of nitrogen supply in the root zone, and the crop growth data does not show nitrogen deficiency, the deduction will be based on the predicted amount of nitrogen release from straw sources.

[0062] The predicted nitrogen and potassium release from straw were determined based on the amount of dry matter returned to the field, the nitrogen and potassium content of straw, and the nutrient release coefficients corresponding to the straw decomposition stage labels for the respective fertilization zones. The nitrogen and potassium content of straw were obtained from straw samples taken before returning to the field. The nutrient release coefficients were set to correspond to the decomposition stage labels and updated based on the field sample database.

[0063] The predicted amount of nitrogen release from straw is determined using the following formula: ; The predicted amount of straw-borne emissions is determined using the following formula: ; in, For the first The fertilization zone is in the first Each sampling node corresponds to the predicted amount of straw nitrogen release during the decomposition stage, in kg / hm; For the first The fertilization zone is in the first Each sampling node corresponds to the predicted amount of potassium release from straw during the decomposition stage, in kg / hm; For the first The amount of dry matter returned to the field by each fertilization zone, in kg / hm; For the first Nitrogen content in straw samples from different fertilization zones; For the first Potassium content in straw samples from different fertilization zones; To be labeled with the straw decomposition stage The corresponding nitrogen release coefficient; To be labeled with the straw decomposition stage The corresponding potassium release coefficient; This indicates one of the following decomposition stages: the decomposition initiation stage, the microbial nitrogen fixation enhancement stage, the slow nitrogen release stage, or the nutrient release compensation stage.

[0064] When implementing the nutrient release coefficient for the first time and before the field sample bank has accumulated sufficient samples, the initial values ​​of the nitrogen and potassium release coefficients are determined from field decomposition test data under similar crop straw types, similar crushing lengths, similar soil penetration depths, and similar soil moisture conditions, local agricultural technical data, or existing test records of the applicant. After completing at least one sampling cycle, the nutrient release coefficients for the corresponding decomposition stage are updated based on the actual measured percentage of residual straw mass, changes in straw nitrogen content, and changes in straw potassium content. Samples from different crop straw types, different returning depths, or significantly different moisture conditions are not combined for calculating the nutrient release coefficients.

[0065] Fertilizer prescriptions are generated on a fertilization zone basis. Each zone's prescription includes the nitrogen, phosphorus, and potassium ratio, application timing, fertilizer form, application depth, and application rate per unit area. Application timings include basal fertilizer application, decomposition initiation compensating fertilizer application, seedling topdressing, rapid vegetative growth topdressing, and pre-yield formation topdressing. Fertilizer forms include fast-acting nitrogen fertilizer, controlled-release nitrogen fertilizer, phosphorus fertilizer, and potassium fertilizer; the specific fertilizer type is determined based on local fertilizer availability, crop type, and soil testing results.

[0066] The nitrogen and fertilizer requirements for each crop stage are not arbitrarily generated by the model, but are determined based on the target crop type, target yield level, local soil testing and fertilizer recommendation indicators, historical yield records of the target farmland, and the corresponding fertilizer requirements during the growth stage. For crops with local fertilizer recommendation procedures, the stage-specific fertilizer requirement ratios in the local recommendation procedures are used first; for crops lacking stage-specific fertilizer requirement ratios, the yield, soil nutrient test values, and post-harvest soil residual nutrient data of similar plots in the previous planting season are used as the initial determination basis, and the data are updated after feedback at the harvest period.

[0067] When generating a fertilization prescription, the basic fertilization requirements are first determined based on the crop's nutrient requirements at each stage and the availability of readily available nutrients in the soil. Then, the allocation of nitrogen fertilizer at different application periods is determined based on the straw decomposition stage label and the nitrogen competition window. Subsequently, the amount of nitrogen fertilizer allocated in the topdressing during the rapid vegetative growth period or the early yield formation period is deducted based on the predicted amount of nitrogen release from straw sources, and the amount of potassium fertilizer applied is deducted based on the predicted amount of potassium release from straw sources. Finally, the theoretical fertilization amount is converted into the fertilization amount per unit area that the equipment can execute, taking into account the fertilization control accuracy of the variable fertilization equipment.

[0068] For fertilization zones in the microbial nitrogen fixation enhancement period and within the nitrogen competition window, the fertilization formula should increase the nitrogen allocation of the decomposition initiation compensation fertilizer or seedling topdressing, and adopt the fertilizer form of quick-acting nitrogen fertilizer or a combination of quick-acting nitrogen fertilizer and controlled-release nitrogen fertilizer. For fertilization zones in the nutrient release compensation period, the fertilization formula should reduce the nitrogen allocation of the topdressing during the rapid vegetative growth period or the early yield formation period, and reduce the amount of potassium fertilizer applied based on the predicted potassium release from straw.

[0069] The variable-rate fertilization equipment generates fertilization control instructions based on the fertilization zone number, fertilizer application rate per unit area, fertilizer form, and fertilization depth in the fertilization formula. The fertilization control instructions include the speed of the fertilizer discharge motor, the opening degree of the fertilizer discharge port, the depth of the furrow opener in the soil, the coordinates of the working path, and the start and end positions of fertilization.

[0070] Variable-rate fertilization equipment undergoes fertilizer discharge calibration before executing the fertilization prescription. During calibration, the fertilizer discharge mass per unit time is collected under different combinations of discharging motor speed, discharging port opening, and operating speed to establish a calibration relationship between discharging motor speed, discharging port opening, operating speed, and fertilizer application rate per unit area. When performing zoned fertilization, the controller determines the discharging motor speed and discharging port opening based on the fertilizer application rate per unit area, fertilizer form, current operating speed, and the aforementioned calibration relationship for the corresponding fertilization zone. The trencher's penetration depth is calibrated using depth sensor detection values ​​or mechanical limiter positions and corresponds to the fertilization depth in the fertilization prescription.

[0071] After the variable-rate fertilization equipment enters the coordinate range of the corresponding fertilization zone, it reads the fertilization control command for that zone and adjusts the speed of the fertilizer discharge motor, the opening of the fertilizer discharge port, and the depth of the furrow opener according to the control command. During fertilization, the variable-rate fertilization equipment records the actual fertilizer discharge volume, fertilization location, fertilization time, operating speed, and changes in the remaining fertilizer in the fertilizer tank. The actual fertilizer discharge volume can be calculated from the fertilizer discharge shaft speed and the opening of the fertilizer discharge port according to the calibration relationship, or it can be obtained from the online weighing module or flow detection module. If there is a deviation between the actual fertilizer discharge volume and the prescription requirements, the deviation is recorded and compensated for within the same fertilization zone or corrected in the next topdressing prescription.

[0072] After fertilization, root zone mineral nitrogen variation data, crop growth feedback data, and harvest yield data were collected at preset sampling nodes for each fertilization zone. Root zone mineral nitrogen variation data included nitrate nitrogen content, ammonium nitrogen content, and the root zone mineral nitrogen content calculated from these values ​​at adjacent sampling nodes before and after fertilization. Crop growth feedback data included leaf color index, canopy coverage, leaf area index, and plant nitrogen nutrition index. Harvest yield data included grain yield, aboveground biomass, and post-harvest soil residual mineral nitrogen data.

[0073] The aforementioned feedback data, along with the actual fertilizer application rate, application location, and application time, are written into the field sample database for the corresponding fertilization zone. After the field sample database is updated, the percentage of residual straw mass, changes in straw nitrogen content, and changes in straw potassium content at different decomposition stages are recalculated, and the nutrient release coefficient is updated. Simultaneously, new samples are added to the training samples of the straw decomposition state recognition model for retraining or incremental updates. The updated model is used for straw decomposition stage identification at the next sampling node or in the next planting season.

[0074] When updating the model, the final yield is not used as the sole basis for model correction. Instead, the actual amount of fertilizer applied, changes in mineral nitrogen in the root zone, changes in crop growth, and the straw decomposition stage labels are used as the basis for updating. This is to avoid mistaking yield fluctuations caused by abnormal climate, pests and diseases, or variety differences for fertilizer prescription errors.

[0075] Taking wheat-corn rotation farmland as an example, straw is crushed and returned to the field after corn harvest. Before operation, the boundary coordinates of the target farmland are obtained, and fertilization zones are divided according to the operating width of the straw returning machinery and the single-discharge control width of the variable fertilizer equipment. The corn harvester records the grain yield, and the system determines the amount of straw dry matter returned to the field in each fertilization zone by combining the corn straw-to-grain ratio and the straw returning ratio. After straw returning, surface images are collected to extract the straw crushing length distribution and straw spreading uniformity; the straw penetration depth and returning layer thickness are determined by tillage depth sensors and soil profile sampling; and straw moisture content, straw carbon-to-nitrogen ratio, straw nitrogen content, and straw potassium content are obtained through straw sample testing.

[0076] Before wheat sowing, the mineral nitrogen, available phosphorus, available potassium, and organic matter contents of soil layers at depths of 010cm, 1020cm, and 2040cm were collected, and the temperature, moisture content, and redox potential of the straw decomposition layer were recorded. After wheat emergence, leaf color index, canopy coverage, leaf area index, and plant nitrogen nutrient index were collected according to fertilization zones. The system inputs the straw decomposition state map, soil stratification nutrient data, straw decomposition environment data, and crop growth data into the straw decomposition state recognition model, and outputs the decomposition stage labels for each fertilization zone.

[0077] For fertilization zones identified as being in the microbial nitrogen fixation enhancement phase, if two consecutive sampling nodes both satisfy the condition that the root zone mineral nitrogen content is below the root zone nitrogen supply limit and the straw carbon-nitrogen ratio is not less than 0.70 times the straw carbon-nitrogen ratio before sowing, then the nitrogen competition window starting node is determined. The system includes the difference between the root zone mineral nitrogen content and the root zone nitrogen supply limit in the nitrogen fertilizer allocation for the decomposition initiation compensation fertilizer or seedling topdressing. For fertilization zones that subsequently transition to the nutrient release compensation phase, the system deducts the nitrogen fertilizer allocation for topdressing during the rapid vegetative growth period or the early yield formation period based on the predicted straw source nitrogen release, and deducts the potassium fertilizer application based on the predicted straw source potassium release.

[0078] The variable-rate fertilization equipment applies fertilizer according to the prescribed ratios for each fertilization zone. During fertilization, the equipment reads the corresponding prescription based on the fertilization zone number and coordinate range, and adjusts the speed of the fertilizer discharge motor, the opening of the fertilizer discharge port, and the depth of the furrow opener. After fertilization, the actual amount of fertilizer discharged, the fertilization location, and the fertilization time are recorded. Subsequently, data on changes in mineral nitrogen in the root zone, crop growth feedback data, and yield data at harvest time are collected at the corresponding sampling nodes. This data is then written into the field sample library to update the straw decomposition status identification model and nutrient release coefficient.

[0079] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine learning-based intelligent method for regulating crop planting by combining fertilizer application with straw return to the field, characterized in that, Includes the following steps: S1. During the straw return operation, the target farmland is divided into multiple fertilization zones. Straw return operation data of each fertilization zone is collected, and a return layer state map of the corresponding fertilization zone is established based on the straw return operation data. The straw return operation data includes the amount of straw returned, the distribution of straw crushing length, the uniformity of straw spreading, the depth of straw entering the soil, the thickness of the return layer, the straw moisture content, and the straw carbon-nitrogen ratio. S2. Before sowing and at the pre-set growth stage of the crop, collect soil stratified nutrient data, return layer environmental data and crop growth data for each fertilization zone. Soil stratified nutrient data includes mineral nitrogen content, available phosphorus content, available potassium content and organic matter content. Return layer environmental data includes return layer temperature, return layer moisture content and return layer redox potential. S3. Input the straw decomposition status map, soil stratification nutrient data, straw decomposition environment data and crop growth data into the straw decomposition status recognition model trained by field samples to obtain the straw decomposition stage label corresponding to each fertilization zone. The straw decomposition stage label includes the decomposition initiation period, the microbial nitrogen fixation enhancement period, the slow nitrogen release period and the nutrient release compensation period. S4. Based on the straw decomposition stage label, root zone mineral nitrogen content, crop stage nitrogen requirement and straw carbon-nitrogen ratio, determine the nitrogen competition window for each fertilization zone, and determine the nitrogen fertilizer allocation for each fertilization zone in basal fertilizer, decomposition start-up compensation fertilizer, seedling topdressing, rapid vegetative growth period topdressing and yield formation early topdressing. Decomposition start-up compensation fertilizer is the nitrogen fertilizer unit applied from straw return to the field to before the seedling stage. S5. Based on the predicted release of nitrogen from straw sources, the predicted release of potassium from straw sources, the deficit of available nutrients in the soil, and the fertilizer requirements of crops at each fertilization zone, generate the corresponding fertilizer formula for the fertilization zone. The fertilizer formula includes the nitrogen-phosphorus-potassium ratio, fertilizer application time, fertilizer form, fertilization depth, and fertilizer application rate per unit area. S6. Apply fertilizer to the target farmland in zones according to the fertilizer formula and control variable fertilization equipment. After fertilization, collect data on the change of mineral nitrogen in the root zone, crop growth feedback data and yield data at harvest time for each fertilization zone. Update the field sample library based on the collected results and retrain the straw decomposition status recognition model using the updated field sample library.

2. The intelligent fertilization and planting regulation method for returning farmland straw to the field based on machine learning as described in claim 1, characterized in that, In S1, when dividing the target farmland into multiple fertilization zones, the boundary coordinates and operation path coordinates of the target farmland are first obtained. Then, the coordinate range of the fertilization zone is determined according to the operation width of the straw returning machine, the single fertilizer discharge control width of the variable fertilizer equipment, and the operation path direction. Each fertilization zone is assigned a zone number. The straw return layer status map includes the fertilization zone number, the coordinate range of the fertilization zone, and the straw return operation data corresponding to the fertilization zone.

3. The intelligent fertilization and planting regulation method for returning farmland straw to the field based on machine learning as described in claim 2, characterized in that, In S1, the amount of straw returned to the field is determined by the harvester output monitoring data, the crop straw-to-grain ratio, and the straw-to-field ratio; The length distribution of the crushed straw was extracted from the surface image after the straw was returned to the field; The uniformity of straw spreading is determined by the ratio of the straw coverage area to the area of ​​the fertilization zone within the same fertilization zone; The straw penetration depth and the thickness of the returned layer are obtained by correcting the values ​​detected by the tillage depth sensor and the soil profile sampling values.

4. The intelligent fertilization and planting regulation method for returning farmland straw to the field based on machine learning as described in claim 3, characterized in that, In S2, the soil nutrient data for each soil layer was collected from the 010cm soil layer, the 1020cm soil layer, and the 2040cm soil layer, respectively. The preset growth period nodes of the crop include pre-sowing, seedling stage, rapid vegetative growth stage, pre-yield formation stage and harvest stage, and the preset growth period nodes of the crop are determined according to the cultivation technical regulations corresponding to the target crop. The crop growth data includes leaf color index, canopy coverage, leaf area index, and plant nitrogen nutrition index.

5. The intelligent fertilization and planting regulation method for returning farmland straw to the field based on machine learning according to claim 4, characterized in that, In S3, the field samples include the return layer state map data, soil stratification nutrient data, return layer environmental data, crop growth data, and straw decomposition stage labels under the same sampling node in the same fertilization zone; The straw decomposition stage labels are based on the percentage of residual straw mass, changes in mineral nitrogen in the returned layer, changes in soluble organic carbon in the returned layer, and changes in the straw carbon-nitrogen ratio. Specifically, the decomposition initiation stage is marked when the percentage of residual straw mass begins to decrease and the soluble organic carbon in the returned layer increases. The microbial nitrogen fixation enhancement stage is marked when the mineral nitrogen content in the root zone continuously decreases and the straw carbon-nitrogen ratio remains no less than 0.70 times the straw carbon-nitrogen ratio before sowing. The slow nitrogen release stage is marked when the percentage of residual straw mass continues to decrease and the mineral nitrogen content in the root zone stops decreasing continuously. The nutrient release compensation stage is marked when the mineral nitrogen content in the root zone reaches the lower limit of nitrogen supply in the root zone and the straw carbon-nitrogen ratio is less than 0.70 times the straw carbon-nitrogen ratio before sowing.

6. The intelligent fertilization and planting regulation method for returning farmland straw to the field based on machine learning according to claim 5, characterized in that, In S3, the straw decomposition state identification model is a classification model, which includes a feature preprocessing module, a classifier, and an output module. The feature preprocessing module standardizes or normalizes the input features, which include straw return amount, straw crushing length distribution, straw penetration depth, return layer temperature, return layer moisture content, return layer redox potential, root zone mineral nitrogen content, available potassium content, leaf color index, and canopy coverage. The classifier uses any one of the following: random forest classifier, gradient boosting tree classifier, or support vector machine classifier. The output module outputs a label for one of the straw decomposition stages: decomposition initiation period, microbial nitrogen fixation enhancement period, slow nitrogen release period, and nutrient release compensation period.

7. The intelligent fertilization and planting regulation method for returning farmland straw to the field based on machine learning as described in claim 6, characterized in that, In S4, the nitrogen requirement of the crop stage is determined based on the target crop type, target yield level, and preset growth period of the crop; The lower limit of nitrogen supply in the root zone is determined based on the nitrogen requirement of the crop stage, the thickness of the soil layer for effective root absorption, soil bulk density, and the area of ​​the fertilization zone. The mineral nitrogen content in the root zone is obtained by adding the nitrate nitrogen content and ammonium nitrogen content in the soil layer where the roots can effectively absorb nitrogen.

8. The intelligent fertilization and planting regulation method for returning farmland straw to the field based on machine learning according to claim 7, characterized in that, In S4, the starting node of the nitrogen competition window is the earlier sampling node when two consecutive sampling nodes in the same fertilization zone meet the following conditions: the straw decomposition stage is labeled as the microbial nitrogen fixation enhancement period, the mineral nitrogen content in the root zone is lower than the lower limit of nitrogen supply in the root zone, and the straw carbon-nitrogen ratio is not lower than 0.70 times the straw carbon-nitrogen ratio before sowing in the fertilization zone. The termination node of the nitrogen competition window is the sampling node when the mineral nitrogen content in the root zone of the fertilization zone reaches the lower limit of nitrogen supply in the root zone, and the straw decomposition stage label changes to the slow nitrogen release period or nutrient release compensation period.

9. A machine learning-based intelligent method for regulating crop straw return and fertilization in farmland, as described in claim 8, is characterized in that... In S4, when the fertilization zone is within the nitrogen competition window, the difference between the mineral nitrogen content in the root zone and the lower limit of nitrogen supply in the root zone is included in the nitrogen fertilizer allocation of the decomposition start-up compensation fertilizer and seedling topdressing. When the fertilization zone is in the nutrient release compensation period, the amount of nitrogen fertilizer allocated for topdressing during the rapid vegetative growth period or the early yield formation period is reduced according to the predicted amount of nitrogen release from straw sources.

10. A machine learning-based intelligent method for regulating crop straw return and fertilization in farmland, as described in claim 9, is characterized in that... In S5, the predicted amount of nitrogen release from straw source and the predicted amount of potassium release from straw source are determined based on the amount of straw dry matter returned to the field, the straw nitrogen content, the straw potassium content, and the nutrient release coefficient corresponding to the straw decomposition stage label of the corresponding fertilization zone. The straw nitrogen content and straw potassium content are obtained by detecting straw samples before returning to the field. The nutrient release coefficient is determined based on the percentage of residual straw mass, the change in straw nitrogen content, and the change in straw potassium content corresponding to the same straw decomposition stage label in the field sample bank. In S6, the variable fertilization equipment generates fertilization control instructions according to the fertilization zone number, fertilizer amount per unit area, fertilizer form and fertilization depth in the fertilization formula, and adjusts the speed of the fertilizer discharge motor, the opening of the fertilizer discharge port and the soil penetration depth of the furrow opener according to the fertilization control instructions.

Images