A system and method for preventing erosion and conserving soil moisture based on straw mulching

Abstract

The present invention discloses a straw mulching-based erosion prevention and moisture conservation system and method, which relates to the field of agricultural technology. The system includes a data acquisition module, a generative adversarial network model module, and an optimization design module. The data acquisition module is used to collect environmental parameter data of the field. The generative adversarial network model module includes a generator and a discriminator. The generator is used to generate a straw mulching scheme based on the environmental parameter data, and the discriminator is used to evaluate the erosion prevention and moisture conservation effect of the straw mulching scheme. The optimization design module is used to generate a final straw mulching optimization design scheme based on the output of the generative adversarial network model module. The present invention solves the problems of uneven mulching, low efficiency, and difficulty in evaluating the effect of traditional straw mulching methods, thereby improving the erosion prevention and moisture conservation effect.

Description

Technical Field

[0001] The present invention relates to the technical field of agriculture, and more particularly to a erosion prevention and moisture conservation system and method based on straw mulching. Background Art

[0002] Straw mulching is a widely used soil conservation technique in agricultural production. By laying crop straw on the surface of fields, it effectively reduces soil erosion caused by wind and water flow, while also conserving soil moisture and improving soil structure. This method has a certain application basis in traditional agriculture and is considered an important means of improving sustainable land use, particularly in arid or windy areas. However, in practice, traditional straw mulching still faces several challenges that need to be addressed. For example, manual or mechanical straw application often results in uneven mulch distribution; some areas may be too thick, affecting crop growth, while others may be too thin, failing to effectively prevent erosion and conserve moisture. Furthermore, this method is inefficient, especially on large-scale farmland. Manual operation is time-consuming, and mechanical equipment lacks the flexibility to adapt to the complex field environment. Furthermore, there is a lack of scientific evidence to evaluate mulching effectiveness, and inconsistent straw decomposition rates can also affect long-term soil conservation. These issues make traditional straw mulching methods insufficient for the precision and intelligent demands of modern agriculture, and improvements are urgently needed to enhance their practicality and effectiveness. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide an anti-corrosion and moisture conservation system and method based on straw mulching, so as to solve the problems mentioned in the background technology.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A straw-mulching-based erosion prevention and moisture conservation system includes a data acquisition module, a generative adversarial network model module, and an optimization design module;

[0006] The data acquisition module is used to collect environmental parameter data of the field;

[0007] The generative adversarial network model module includes a generator and a discriminator, wherein the generator is used to generate a straw mulching scheme according to the environmental parameter data, and the discriminator is used to evaluate the anti-corrosion and moisture conservation effect of the straw mulching scheme;

[0008] The optimization design module is used to generate a final straw covering optimization design scheme based on the output of the generative adversarial network model module.

[0009] In some embodiments, the generator adopts a convolutional neural network architecture combined with a Transformer module to generate a cover thickness distribution map and a cover density distribution map; the discriminator adopts a multi-task learning architecture to simultaneously evaluate the soil erosion reduction rate and moisture retention rate of the straw covering scheme.

[0010] In some embodiments, the following loss function is used during the training of the generative adversarial network model module:

[0011] ;

[0012] in, is the total loss, For data-driven losses, is the loss based on the physical model, and is the weight coefficient;

[0013] The data-driven loss is the difference between the straw mulching scheme generated by the generator and the historical high-quality mulching scheme, and the data comes from the historical straw mulching scheme records;

[0014] The loss of the physical model is the difference between the soil erosion amount under the generation scenario calculated according to the universal soil loss equation and the target erosion amount. The physical model is a universal soil loss equation for evaluating soil erosion and water conservation in the agricultural field.

[0015] In some embodiments, the environmental parameter data includes any one or more of the following: soil type data, terrain slope data, rainfall distribution data, and vegetation coverage data.

[0016] In some embodiments, the data acquisition module further includes a wind flow field data acquisition unit for collecting wind direction and wind speed data of the field; the generator is further configured to generate straw morphology recommendations based on the wind flow field data in response to wind force influences.

[0017] In some embodiments, the optimization design module further includes a morphology optimization submodule and a thickness optimization submodule;

[0018] The morphology optimization submodule is used to recommend the optimal straw morphology based on wind flow field data;

[0019] The thickness optimization submodule is used to optimize the cover thickness according to the terrain slope and soil type.

[0020] In some embodiments, the morphology optimization submodule adopts a stability evaluation model, which calculates the wind resistance stability of the straw morphology based on the following formula:

[0021] ;

[0022] in:

[0023] is the stability index;

[0024] is the number of sampling points in the wind flow field;

[0025] For the Wind speed at each sampling point;

[0026] is the maximum wind speed;

[0027] For the Recommended straw morphology for each sampling point;

[0028] For the Wind direction angle at each sampling point;

[0029] For form In wind direction The stability function below.

[0030] The present invention also discloses a method for preventing erosion and conserving soil moisture based on straw mulching, comprising the following steps:

[0031] Collect environmental parameter data of the field through the data acquisition module;

[0032] Inputting the environmental parameter data into a generative adversarial network model module, the generative adversarial network model module includes a generator and a discriminator, wherein the generator is used to generate a straw mulching scheme according to the environmental parameter data, and the discriminator is used to evaluate the anti-corrosion and moisture conservation effect of the straw mulching scheme;

[0033] Based on the output results of the generative adversarial network model module, the optimization design module generates an optimized design scheme for straw mulching.

[0034] In some embodiments, when collecting environmental parameter data, wind flow field data of the field is further collected, including wind speed and wind direction;

[0035] The optimization design module includes a morphology optimization submodule, and the morphology optimization submodule recommends an optimal straw morphology based on wind flow field data.

[0036] In some embodiments, the method further includes a training process for a generative adversarial network model, wherein the following loss function is used during the training process:

[0037] ;

[0038] in, is the total loss, For data-driven losses, is the loss based on the physical model, and is the weight coefficient;

[0039] The data-driven loss is the difference between the straw mulching scheme generated by the generator and the historical high-quality mulching scheme, and the data comes from the historical straw mulching scheme records;

[0040] The loss of the physical model is the difference between the soil erosion amount under the generation scenario calculated according to the universal soil loss equation and the target erosion amount. The physical model is a universal soil loss equation for evaluating soil erosion and water conservation in the agricultural field.

[0041] The advantage of the present invention over the prior art is that, by introducing intelligent design methods, the present invention significantly improves the problems of uneven coverage and low efficiency in traditional straw mulching. With the collaborative work of the data acquisition module, the generative adversarial network model module and the optimization design module, the system can generate personalized mulching plans based on the specific environment of the field, which not only improves the uniformity of the coverage, but also greatly improves the implementation efficiency, thereby enhancing the overall effect of erosion prevention and moisture conservation. On this basis, further innovations make the plan more perfect. By combining the wind flow field data of the field and a variety of environmental parameters, such as soil type, terrain slope, etc., the system can recommend the optimal straw shape and mulching thickness distribution. This design improves the adaptability of the mulching plan to complex natural conditions, enhances the stability of the straw under the action of wind, and optimizes the effects of water retention and soil erosion control, providing more reliable technical support for the sustainable development of agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is the overall flow chart of the present invention;

[0043] Figure 2 Schematic diagram of the training process of the generative adversarial model of the present invention;

[0044] Figure 3 It is a schematic diagram of the present invention considering the air outlet situation. DETAILED DESCRIPTION

[0045] The specific embodiments of the present invention will be described below with reference to the accompanying drawings.

[0046] The system consists of three main parts: a data acquisition module, a generative adversarial network model module, and an optimization design module. These parts work closely together to ensure that the entire process from data acquisition to solution generation is scientific and efficient.

[0047] The data acquisition module is responsible for acquiring environmental parameter data for the fields, which serves as the foundation for subsequent plan generation. Using sensors, drone aerial photography, or satellite remote sensing, it can collect a variety of information, including soil type, terrain slope, rainfall distribution, and vegetation coverage. For example, soil type data reflects the physical properties of the field soil, such as sand or clay, while terrain slope data reveals the undulations of the field, helping to determine coverage requirements for different areas.

[0048] The generative adversarial network model consists of two sub-parts: the generator and the discriminator, which are the main bodies of the generative coverage method.

[0049] As shown in Figure 1, the generator uses collected environmental parameter data to generate straw mulching plans for specific fields, including mulch thickness and density distribution maps. The discriminator evaluates these plans, analyzing their effectiveness in preventing wind erosion and conserving soil moisture, ensuring that the generated plans have practical application value.

[0050] To enable the generator to handle complex field data, it uses a convolutional neural network architecture combined with a Transformer module. Convolutional neural networks excel at processing spatially distributed data and can generate accurate coverage distribution maps. The Transformer module, through its self-attention mechanism, captures the correlations between different areas of the field, improving the overall coordination of the solution.

[0051] The discriminator uses a multi-task learning architecture to simultaneously evaluate the solution's soil erosion reduction and water conservation. For example, on a steeply sloping field, the generator might recommend increasing the thickness of mulch at the bottom of the slope to reduce water runoff, while the discriminator verifies whether this adjustment also improves water conservation.

[0052] An optimization design module can be further introduced to combine the intelligent generation results of the generative adversarial network (GAN) with agricultural practice to produce scientific and practical solutions. The optimization design module acts as a "solution improver" in the system. Based on the preliminary solution generated by the GAN and the evaluation results of the discriminator, it adjusts parameters such as mulch thickness and density to make the solution more consistent with actual field conditions (such as topography, soil type, and resource availability).

[0053] The optimization design module can be implemented using rules and algorithms. The specific process includes: first, receiving the initial plan from the GAN generator and feedback from the discriminator; then, adjusting parameters based on rules developed by agricultural experts (such as increasing mulch thickness for steep slopes) and actual field conditions; optimizing the plan based on resource constraints to ensure feasibility; and finally, iteratively adjusting the plan to ensure that it meets quality standards while adapting to actual operational needs. This approach combines the intelligence of GANs with human experience, improving the practicality and effectiveness of the plan. In some embodiments, adjustments can be made through both manual adjustments and pre-defined rules. For example, a GAN may generate a plan with a complex mulch thickness distribution. However, given the limited resources and labor of local farmers, the optimization design module can appropriately simplify it, such as by evenly reducing the thickness, to make it easier for farmers to implement. These rules can be manually defined.

[0054] As shown in Figure 2, during training, the generative adversarial network model uses a hybrid loss function to guide model optimization. The specific form is as follows:

[0055] ;

[0056] in, is the total loss, For data-driven losses, is the loss based on the physical model, and is the weight coefficient;

[0057] The selection of λ1 and λ2 can be based on the numerical range of the two types of losses during model training and the weight preference of the actual focus:

[0058] If you pay more attention to the consistency between the solution and historical high-quality samples (emphasizing experience matching), you can set: λ1 = 0.7, λ2 = 0.3;

[0059] If we attach more importance to the rationality and scientific nature of the physical model (emphasizing theoretical constraints), we can set: λ1 = 0.4, λ2 = 0.6;

[0060] If you want to balance the two, it is generally set to: λ1 = λ2 = 0.5;

[0061] In practice, the optimal combination can be selected through cross-validation, such as grid search in the range of [0.1, 0.9].

[0062] The data-driven loss measures the difference between the straw mulching solutions generated by the generator and historically proven good mulching solutions. These historical good mulching solutions are derived from successful agricultural practices and include records of parameters such as mulch thickness and density. By calculating the deviation of the generated solutions from these good solutions (typically using mean squared error), the model can learn empirically validated effective patterns.

[0063] More specifically, suppose the generator generates a coverage plan where The coverage thickness of each sampling point is , the coverage density is ;

[0064] The corresponding coverage thickness in the historical high-quality coverage plan is , the coverage density is ;

[0065] is the total number of sampling points in the scheme, then The specific formula can be:

[0066] ;

[0067] in:

[0068] It represents the square of the deviation between the generated solution and the historical solution in terms of coverage thickness;

[0069] represents the square of the deviation in coverage density;

[0070] Taking the average Ensure that the loss is independent of the number of sampling points and is universal.

[0071] In this way, the model is able to learn effective cover patterns empirically verified in historical data, such as combinations of cover thickness and density that perform well in specific areas.

[0072] Regarding the loss of the physical model, the soil erosion amount under the generated scenario is calculated based on the universal soil loss equation and compared with the target erosion amount. The universal soil loss equation is a classic model for evaluating soil erosion and water conservation in the agricultural field, which can provide a scientific physical basis for the scenario. The weight coefficient is used to balance the contribution of the data-driven and physical model losses. By adjusting these coefficients, the training direction can be optimized according to the specific needs of the field (such as paying more attention to erosion prevention or moisture conservation). For example, in areas with heavy rainfall, the training may increase to ensure that the mulching scheme performs better in reducing soil erosion.

[0073] More specifically, the Universal Soil Loss Equation is a model widely used in agriculture to assess soil erosion. Its basic form is:

[0074] ;in:

[0075] A: average annual soil erosion (unit: tons / hectare / year);

[0076] R: rainfall erosivity factor, reflecting the impact of rainfall on soil erosion;

[0077] K: soil erodibility factor, which indicates the sensitivity of soil to erosion;

[0078] L: slope length factor, related to the slope length;

[0079] S: slope factor, related to the steepness of the slope;

[0080] C: vegetation cover and management factor, reflecting the control effect of vegetation or cover on soil erosion;

[0081] P: Soil and water conservation measures factor, which indicates the effectiveness of artificial measures (such as straw mulching) on ​​erosion prevention and control.

[0082] In the straw covering scenario of the present invention, the covering scheme generated by the generator mainly affects factors (by changing land cover) and Factor (by improving soil and water conservation effect). Assume that the factor corresponding to the coverage scheme generated by the generator is and , the factor of the target coverage scheme (based on expert experience or ideal conditions) is and .but:

[0083] The amount of soil erosion under the generated scenario is: ;

[0084] The amount of soil erosion under the target scenario is: ;

[0085] The loss function of the physical model is defined as the square of the difference between the generated solution soil erosion and the target erosion: ;

[0086] To simplify the calculations and highlight the impact of the coverage scheme, it can be assumed that are constants (these factors are usually determined by environmental conditions and remain constant within the same field). Thus:

[0087] ;

[0088] pass Training can guide the generator to produce a coverage scheme that conforms to physical laws and makes the soil erosion amount as close to the target value as possible.

[0089] As can be seen above, environmental parameter data includes soil type, terrain slope, rainfall distribution, and vegetation cover. Soil type data reveals the soil's water retention capacity and erosion propensity. For example, sandy soil requires denser cover to prevent water loss. Terrain slope data helps determine the required cover thickness in different areas. Rainfall distribution data reflects the water input to the field, while vegetation cover provides a reference for natural erosion protection. This data can be collected through soil sensors, topographic survey instruments, weather stations, or remote sensing imagery.

[0090] As shown in Figure 3, in some embodiments, to improve the adaptability of mulching solutions to wind conditions, the data acquisition module also includes a wind flow data acquisition unit for recording wind direction and speed data in the field. By deploying wind direction and speed recorders, a real-time wind flow distribution map can be generated. This data provides additional input to the generator, enabling it to recommend appropriate straw shapes, such as strips, crumbs, or curls, based on wind influences to enhance mulching stability.

[0091] In another embodiment, when generating the final solution, the optimization design module is further subdivided into a shape optimization submodule and a thickness optimization submodule to respectively address the optimization requirements of straw shape and covering thickness. The shape optimization submodule recommends the optimal straw shape based on wind flow field data. To quantify the wind resistance stability of different shapes, this submodule uses a stability assessment model, and the calculation formula is as follows:

[0092] ;

[0093] in:

[0094] is the stability index, the larger the value, the more stable the morphology;

[0095] is the number of sampling points in the wind flow field;

[0096] For the Wind speed at each sampling point;

[0097] is the maximum wind speed;

[0098] For the Recommended straw morphology for each sampling point;

[0099] For the Wind direction angle at each sampling point;

[0100] For form In wind direction The stability function under the condition is usually determined by experiments or simulations.

[0101] For example, in areas with higher wind speeds, curly straw may be recommended due to its stronger adhesion and thus a higher stability index.

[0102] The thickness optimization submodule adjusts mulch thickness based on terrain slope and soil type. For example, in areas with steep slopes, increasing mulch thickness can effectively slow water flow and reduce erosion risk; in areas with loose soil, adjusting mulch density can improve moisture retention. This zoning optimization approach ensures that mulch solutions can adapt to the diverse characteristics of the fields.

[0103] Correspondingly, the present invention provides a method for preventing erosion and conserving soil moisture based on straw mulching, and the specific implementation steps are as follows:

[0104] First, the data acquisition module collects field environmental parameter data, including soil type, terrain slope, rainfall distribution, and vegetation coverage. If further optimization of wind resistance is required, wind flow field data, such as wind speed and direction, will also be collected.

[0105] This data is then fed into a generative adversarial network model. The generator generates preliminary straw mulching plans based on the input data, including a mulch thickness distribution map, a mulch density distribution map, and possible straw shape recommendations. The discriminator evaluates these plans, analyzing their performance in terms of erosion prevention and moisture conservation, and provides feedback to the generator for further improvement.

[0106] Finally, the optimization design module generates a final straw mulch optimization design based on the output of the generative adversarial network. This design not only includes thickness and density distribution, but also, through the morphology optimization submodule, recommends the optimal straw shape to cope with wind impacts.

[0107] To more intuitively demonstrate the application effects of the present invention, the following example uses a specific field as an example. This field is located in a windy area, with a terrain slope between 10° and 20° and predominantly sandy soil. Traditional mulching methods use a uniform thickness of straw in a single form, resulting in a mulch layer that is easily blown away by the wind, limiting its effectiveness in preventing erosion and conserving moisture.

[0108] After using this system, sensors and wind direction and speed recorders first collected data, indicating an average wind speed of 5 m / s, with winds mostly from the northwest. Based on this information, the generator generated a zoned mulching plan: a mulch thickness of 3 cm at the top of the slope and 5 cm at the bottom; curly straw was recommended for windy areas, while strip straw was used in flat areas. After evaluation by the discriminator, this plan achieved an 85% reduction in soil erosion and a 90% water retention rate, significantly outperforming traditional methods. The morphology optimization submodule further calculated a stability index of 0.75 for curly straw, higher than the 0.50 for strip straw, validating the recommendation.

[0109] After implementation, soil erosion in the field was reduced by 30%, water retention increased by 20%, and the stability of the cover layer was significantly improved. This result demonstrates that the present invention can generate scientific and efficient cover solutions based on the specific conditions of the field.

[0110] By combining generative adversarial network technology with agricultural practice, this invention enables intelligent and personalized design of straw mulching solutions. The system leverages environmental parameters and wind flow data to optimize mulch thickness, density, and morphology. Through the dual constraints of data-driven and physical models, it ensures the scientific and practical nature of the solution. This technology not only improves erosion prevention and soil moisture conservation in fields but also provides farmers with simple, effective mulching guidance, promising broad application prospects.

[0111] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A straw-mulching-based erosion prevention and soil moisture conservation system, characterized in that: Includes data acquisition module, generative adversarial network model module and optimization design module; The data acquisition module is used to collect environmental parameter data of the field; The generative adversarial network model module includes a generator and a discriminator, wherein the generator is used to generate a straw mulching scheme according to the environmental parameter data, and the discriminator is used to evaluate the anti-corrosion and moisture conservation effect of the straw mulching scheme; The optimization design module is used to generate a final straw mulching optimization design scheme according to the output of the generative adversarial network model module; The data acquisition module further includes a wind flow field data acquisition unit for collecting wind direction and wind speed data of the field; the generator is further used to generate straw shape recommendations based on the wind flow field data; The generator uses a convolutional neural network architecture combined with a Transformer module to generate a mulch thickness distribution map and a mulch density distribution map; the discriminator uses a multi-task learning architecture to simultaneously evaluate the soil erosion reduction rate and water retention rate of the straw mulching scheme; The optimization design module also includes a morphology optimization submodule and a thickness optimization submodule; The morphology optimization submodule is used to recommend the optimal straw morphology based on wind flow field data; The thickness optimization submodule is used to optimize the cover thickness according to the terrain slope and soil type; The morphology optimization submodule adopts a stability evaluation model, which calculates the wind resistance stability of the straw morphology based on the following formula: ； in: is the stability index; is the number of sampling points in the wind flow field; For the Wind speed at each sampling point; is the maximum wind speed; For the Recommended straw morphology for each sampling point; For the Wind direction angle at each sampling point; For form In wind direction The stability function below.

2. The anti-corrosion and moisture conservation system based on straw covering according to claim 1 is characterized in that: During the training process of the generative adversarial network model module, the following loss function is used: ； in, is the total loss, For data-driven losses, is the loss based on the physical model, and is the weight coefficient; The data-driven loss is the difference between the straw mulching scheme generated by the generator and the historical high-quality mulching scheme, and the data comes from the historical straw mulching scheme records; The loss of the physical model is the difference between the soil erosion amount under the generation scenario calculated according to the universal soil loss equation and the target erosion amount. The physical model is a universal soil loss equation for evaluating soil erosion and water conservation in the agricultural field.

3. The anti-corrosion and moisture conservation system based on straw covering according to claim 1 is characterized in that: The environmental parameter data includes any one or more of the following: soil type data, terrain slope data, rainfall distribution data, and vegetation coverage data.

Images