Forage grass planting humidity control method

Abstract

The invention relates to the technical field of forage grass planting management, and particularly discloses a forage grass planting humidity control method, which adopts an image processing technology based on deep learning to process a forage grass growth state image so as to extract multi-scale semantic features of a forage grass growth state. And feature significance enhancement of pixel granularity is carried out on the pasture growth state features, so that the feature distinction degree is improved, finer pasture growth state information is captured, and the pasture growth state tag is intelligently identified. And on the basis, a soil humidity and an irrigation strategy which are most suitable for the current growth state of the pasture are provided for the pasture by further combining a pasture growth knowledge graph. In this way, it can be ensured that forage grass obtains appropriate water supply in different growth stages.

Description

Technical Field

[0001] The present application relates to the technical field of forage planting management, and more specifically, to a forage planting humidity control method. Background Art

[0002] With the modernization of global agriculture, precision agriculture and smart agriculture have gradually become new trends in agricultural development. As an important basic resource for animal husbandry, the growth status of forage directly affects the production efficiency and economic benefits of animal husbandry. In the process of forage cultivation, soil moisture is one of the key factors affecting forage growth. Appropriate soil moisture can not only promote the growth of forage, improve yield and quality, but also effectively save water resources and reduce growth disorders caused by over-irrigation or drought.

[0003] Traditionally, soil moisture management for forage cultivation mainly relies on manual experience and judgment. However, this method is highly subjective and uncertain, and is not only inefficient, but also difficult to adapt to soil moisture changes in different regions and seasons.

[0004] With the development of agricultural information technology, some humidity control systems based on sensors and the Internet of Things have begun to be used in forage cultivation. By real-time monitoring of soil moisture and automatically adjusting irrigation volume, the accuracy and efficiency of management have been improved. However, most existing humidity control systems are only adjusted based on the physical parameters of soil moisture, triggering irrigation operations based on fixed humidity thresholds, and lack the ability to directly perceive and respond to the growth status of forage. Since the growth status of forage is a complex biological process involving multiple growth stages, it is difficult to fully and accurately assess its actual needs based on soil moisture data alone, which may lead to excessive or insufficient soil moisture.

[0005] Therefore, an optimized method for controlling humidity in forage cultivation is desired. Summary of the invention

[0006] In order to solve the above-mentioned technical problems, the present application is proposed. An embodiment of the present application provides a forage planting humidity control method, which uses deep learning-based image processing technology to process the forage growth status image to extract the multi-scale semantic features of the forage growth status, and enhances the feature significance of the forage growth status features at the pixel granularity to improve the feature discrimination, capture more detailed forage growth status information, and thus intelligently identify the growth status label of the forage. On this basis, it is further combined with the forage growth knowledge graph to provide the forage with the most suitable soil moisture and irrigation strategy for its current growth state. In this way, it can ensure that the forage obtains appropriate water supply at different growth stages.

[0007] According to one aspect of the present application, a forage grass planting humidity control method is provided, which comprises:

[0008] Acquire the grass growth status image collected by the camera;

[0009] Extracting grass growth state features of the grass growth state image to obtain a grass growth state multi-scale semantic coding feature map;

[0010] Performing feature saliency differentiation and enhancement based on pixel granularity on the multi-scale semantic coding feature map of the forage growth state to obtain a multi-scale semantic enhancement coding feature map of the forage growth state;

[0011] Based on the multi-scale semantic enhancement coding feature map of the forage growth state, identifying the forage growth state label;

[0012] Based on the forage growth status label and the forage growth knowledge graph, an irrigation strategy is determined.

[0013] Compared with the prior art, the forage planting humidity control method provided by the present application uses deep learning-based image processing technology to process the forage growth status image to extract the multi-scale semantic features of the forage growth status, and enhances the feature significance of the forage growth status features at the pixel granularity to improve the feature distinction, capture more detailed forage growth status information, and thus intelligently identify the growth status label of the forage. On this basis, it is further combined with the forage growth knowledge map to provide the forage with the most suitable soil moisture and irrigation strategy for its current growth state. In this way, it can ensure that the forage obtains appropriate water supply at different growth stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0015] Figure 1 This is a flow chart of a forage planting humidity control method according to an embodiment of the present application.

[0016] Figure 2 Schematic diagram of data flow of a forage planting humidity control method according to an embodiment of the present application.

[0017] Figure 3 This is a flowchart of sub-step S3 in the forage planting humidity control method according to an embodiment of the present application.

[0018] Figure 4 This is a flowchart of sub-step S32 in the forage planting humidity control method according to an embodiment of the present application.

[0019] Figure 5 This is a flowchart of sub-step S5 in the forage planting humidity control method according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0021] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0022] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0024] It is worth noting that in this application, all actions to obtain data are carried out in compliance with the relevant data protection laws and policies of the country where the data is located, and with the authorization given by the owner of the corresponding device.

[0025] In response to the technical problems described in the above background technology, this application proposes an optimized forage planting humidity control method, which uses deep learning-based image processing technology to process forage growth status images to extract multi-scale semantic features of forage growth status, and enhances the feature significance of forage growth status features at pixel granularity to improve feature discrimination, capture more detailed forage growth status information, and thus intelligently identify forage growth status labels. On this basis, it is further combined with the forage growth knowledge graph to provide forage with the most suitable soil moisture and irrigation strategy for its current growth status. In this way, it can ensure that forage obtains appropriate water supply at different growth stages.

[0026] Figure 1 This is a flow chart of a forage planting humidity control method according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the forage planting humidity control method according to the embodiment of the present application. Figure 1 and Figure 2 As shown, the forage planting humidity control method includes the following steps: S1, acquiring a forage growth status image collected by a camera; S2, extracting the forage growth status features of the forage growth status image to obtain a multi-scale semantic coding feature map of the forage growth status; S3, performing feature significance distinction enhancement based on pixel granularity on the multi-scale semantic coding feature map of the forage growth status to obtain a multi-scale semantic enhancement coding feature map of the forage growth status; S4, identifying a forage growth status label based on the multi-scale semantic enhancement coding feature map of the forage growth status; S5, determining an irrigation strategy based on the forage growth status label and the forage growth knowledge map.

[0027] In the above-mentioned forage planting humidity control method, the step S1 is to obtain the forage growth state image collected by the camera. It should be understood that the use of image technology to monitor the forage growth state is a non-invasive monitoring method, which will not cause physical damage to the forage, and can effectively obtain information related to its growth state without interfering with its natural growth process, thereby providing a reliable basis for the identification and classification of the forage growth stage and the recommended control of soil moisture.

[0028] In the above-mentioned forage planting humidity control method, using a camera to collect images of forage growth status is an important component. This process provides basic data for subsequent image processing, feature extraction and decision making. The specific implementation method of this process will be described in detail below:

[0029] First of all, in response to the needs of monitoring the growth status of forage, it is necessary to choose the right type of camera. Multispectral cameras can capture information in different bands, which helps to more accurately assess the health of vegetation. For example, by analyzing data in the near-infrared band, you can better understand the moisture content and chlorophyll levels of plants, which is crucial to determining whether the forage is healthy. You can choose a standard high-resolution visible light camera, which is low-cost, easy to maintain, and can meet the image acquisition needs in most cases. Considering the environmental challenges of outdoor use, the camera should have good waterproof and dustproof performance (such as IP67 and above), and support Wi-Fi, 4G / 5G or other wireless communication protocols for remote monitoring and data transmission. In addition, the camera should be able to adapt to different network environments to ensure stable upload of image data even when the network signal is weak.

[0030] The camera should be placed at an appropriate height, usually about 2-3 meters from the ground, so that it can cover the entire monitoring area without losing details due to being too high. In order to ensure that the shooting range is as wide as possible and there are no blind spots, it may be necessary to arrange multiple cameras reasonably according to the shape and size of the field, and synthesize a panoramic image through stitching technology to ensure that every corner can be effectively monitored. The angle of the camera should be adjusted to shoot vertically downward to reduce the deformation error caused by the tilt of the viewing angle. For places with complex terrain or obstructions that affect the line of sight, it may be necessary to install additional auxiliary equipment, such as reflectors or increase the height to improve shooting conditions. In order to ensure the stability of the camera, it is necessary to use a stable bracket to fix the camera in a predetermined position to avoid displacement caused by wind and sun that affects the shooting effect. At the same time, considering the impact of agricultural activities on the camera, it is also very important to ensure that it will not be disturbed or damaged by humans.

[0031] The design of shooting frequency requires scientific planning. It should not be too frequent to increase the data volume burden, nor should it be too sparse to miss important change nodes. Generally, the shooting interval can be adjusted according to the growth cycle and seasonal characteristics of the grass. For example, the rapid growth period can be shortened to once or even multiple times a day, and the slow growth period can be extended to several times a week. In addition to timed shooting, trigger shooting can also be combined with specific events detected by the sensor (such as after rain, before and after irrigation) to capture the growth under special conditions. This flexibility can help better understand the response of grass under different environmental conditions, so as to formulate more accurate management strategies. For example, when the soil moisture sensor detects insufficient water, the camera is immediately started to shoot, and the comparison images before and after the grass is dehydrated can be obtained in time, providing a basis for subsequent irrigation decisions. In order to simplify the implementation method, a fixed shooting schedule is set, such as shooting once in the morning and evening every day, so that a relatively stable growth status record can be obtained without relying on a complex trigger mechanism.

[0032] In order to ensure the quality of the image, it is particularly important to pay attention to parameter settings such as white balance correction and exposure compensation to adapt to different lighting conditions and maintain color accuracy. Since changes in natural light will affect the consistency of the image, especially in the morning and evening and on cloudy days, this effect can be mitigated by adding fill light equipment or using a camera with automatic gain control (AGC) function. Regularly cleaning dust and stains on the lens surface is also an essential maintenance work to ensure that the image clarity is not affected. Pre-processing operations such as denoising and contrast enhancement of the original image can be completed through simple image editing software to further improve the effectiveness of subsequent analysis.

[0033] Changes in natural light, extreme weather such as wind, rain and snow, and activities of birds, insects, etc. may interfere with normal shooting. To this end, the impact of lighting changes can be reduced by adding fill light equipment or using cameras with automatic gain control (AGC) function; select weather-resistant products and take necessary protective measures to cope with bad weather; at the same time, set appropriate shooting ranges, angles, and post-filtering algorithms to reduce interference caused by biological activities. For example, in windy areas, choose a camera bracket with a wind-resistant design, or set up a protective net around the camera to prevent foreign objects from invading the lens. For locations that are prone to attracting birds, install bird repellers or take other physical methods to reduce bird interference. In addition, in order to deal with image quality problems that may be caused by long-term continuous shooting, develop an automated image quality inspection system to monitor image quality in real time and issue an alarm when necessary to remind maintenance personnel to check and adjust.

[0034] In summary, a standard high-resolution visible light camera can be selected, which is not only low-cost and easy to maintain, but also meets the image acquisition requirements in most cases. The installation position of the camera needs to consider the height and angle to ensure that the entire monitoring area is covered while avoiding deformation errors. The reasonable arrangement of multiple cameras and the synthesis of panoramic images through stitching technology can ensure that there is no blind spot in the monitoring. The shooting frequency can be flexibly adjusted according to the growth cycle of forage grass. A fixed schedule such as shooting once in the morning and evening every day can provide a stable growth status record without relying on a complex trigger mechanism. In terms of image quality control, the image clarity and consistency can be ensured by manually adjusting the white balance and exposure compensation settings, and regularly maintaining the lens cleaning. In the face of environmental factors, choose a camera and bracket with strong weather resistance, and take physical methods such as protective nets and bird repellers to reduce interference. In this way, images of the growth status of forage grass can be effectively collected, laying a solid foundation for subsequent image analysis.

[0035] In the above-mentioned forage planting humidity control method, the step S2 extracts the forage growth state features of the forage growth state image to obtain a multi-scale semantic coding feature map of the forage growth state. In a specific example of the present application, the step S2 includes: inputting the forage growth state image into a forage growth state feature extractor based on the FPT model to obtain the multi-scale semantic coding feature map of the forage growth state. It should be understood that when processing image data, the traditional convolutional neural network (CNN) mainly relies on local convolution operations to extract local features, while ignoring the contextual interaction between image features. With the introduction of the Transformer architecture in the field of computer vision, the model can better capture the global dependencies in the image, but because the spatial interaction of the Transformer architecture is usually based on a fixed input resolution, it is impossible to capture the non-local context of objects (or parts) at different scales. Therefore, in order to fully extract the multi-scale semantic features of the forage growth state, the present application adopts the FPT (Feature Pyramid Transformer) model to process the forage growth state image. The FPT model combines the Feature Pyramid Network (FPN) with the Transformer, uses the FPN structure to extract the multi-scale feature representation of the grass growth state image, and introduces the Transformer's self-attention mechanism to enhance the interaction between multi-scale features, thereby realizing cross-space and cross-scale grass growth state feature interaction, and obtaining a multi-scale semantic encoding feature map of grass growth state. In this way, the FPT model can more comprehensively understand the complexity of grass growth state and capture key growth state information at different scales, thereby achieving comprehensive perception and accurate description of grass growth state.

[0036] In the above-mentioned forage planting humidity control method, the step S3 performs feature significance differentiation and enhancement based on pixel granularity on the multi-scale semantic coding feature map of forage growth status to obtain a multi-scale semantic enhanced coding feature map of forage growth status. It should be understood that the present application takes into account that in practical applications, the forage growth status image captured by the camera often contains a lot of unnecessary background information or noise, which in turn interferes with the final analysis results. Therefore, in order to more effectively emphasize the feature representation that has important characterization significance for the forage growth status, while removing or weakening irrelevant information, the present application further introduces feature significance enhancement technology, by performing feature significance recognition and hierarchical modulation processing on the multi-scale semantic coding feature map of forage growth status at pixel granularity, so as to enhance the distinguishability and importance of features, effectively highlight the key features of forage growth, and suppress background noise and interference information. Among them, Figure 3FIG. 1 is a flow chart of sub-step S3 in the forage planting humidity control method according to an embodiment of the present application. Figure 3 As shown, the step S3 includes the steps of: S31, performing pixel granularity feature decoupling on the multi-scale semantic coding feature map of forage growth status to obtain a set of pixel granularity local decoupled semantic feature vectors of forage growth status; S32, extracting the semantic significance distribution pattern of the set of pixel granularity local decoupled semantic feature vectors of forage growth status to obtain a fine-grained semantic significance distribution matrix of forage growth status; S33, based on the fine-grained semantic significance distribution matrix of forage growth status, performing feature explicit hierarchical modulation on the multi-scale semantic coding feature map of forage growth status to obtain the multi-scale semantic enhanced coding feature map of forage growth status.

[0037] Specifically, the step S31 is expressed by the formula:

[0038] decouple(Fi)={v1,v2,...,v i ,...,v n}

[0039] Among them, F i represents the multi-scale semantic coding feature map of the forage growth state, decouple(· represents the pixel granularity feature decoupling, v1, v2, v i and v n They respectively represent the first, second, i-th and n-th pixel-granularity local decoupling semantic feature vectors of grass growth status, and n is the number of feature vectors in the set of pixel-granularity local decoupling semantic feature vectors of grass growth status.

[0040] That is, a pixel-granularity feature decoupling operation is performed on the multi-scale semantic coding feature map of the grass growth status to decompose it into a series of independent pixel-granularity local decoupled semantic feature vectors of the grass growth status, so as to eliminate unnecessary coupling relationships between pixels and improve the ability to understand the characteristics of the grass growth status at each local position.

[0041] Figure 4 FIG. 4 is a flowchart of sub-step S32 in the forage planting humidity control method according to an embodiment of the present application. Figure 4As shown, the step S32 includes: S321, inputting each of the grass growth state pixel granularity local decoupling semantic feature vectors in the set of the grass growth state pixel granularity local decoupling semantic feature vectors into a hyperbolic space mapper to obtain a set of hyperbolic space modulated grass growth state pixel granularity local decoupling semantic feature vectors; S322, calculating the fine-grained semantic saliency measurement coefficient of each of the hyperbolic space modulated grass growth state pixel granularity local decoupling semantic feature vectors in the set of the hyperbolic space modulated grass growth state pixel granularity local decoupling semantic feature vectors to obtain the grass growth state fine-grained semantic saliency distribution matrix formed by arranging multiple grass growth state fine-grained semantic saliency measurement coefficients. Wherein, the hyperbolic space mapper is used to perform continuous linear transformation on the grass growth state pixel granularity local decoupling semantic feature vector using a first weight parameter matrix and a second weight parameter matrix to obtain the hyperbolic space modulated grass growth state pixel granularity local decoupling semantic feature vector.

[0042] More specifically, the step S321 is expressed by the formula:

[0043] h i =W1v i W2

[0044] Where W1 and W2 represent the first weight parameter matrix and the second weight parameter matrix respectively, h i Represents the pixel-granular local decoupled semantic feature vector of the i-th hyperbolic spatial modulation grass growth state.

[0045] Here, considering that the hyperbolic geometric space has better adaptability than the Euclidean space when processing hierarchical data sets, the present application further sends the obtained set of local decoupled semantic feature vectors of the pixel granularity of the grass growth status into the hyperbolic space mapper for domain transformation, and maps the grass growth status features from the Euclidean geometric space to the hyperbolic geometric space, so as to utilize the characteristics of the hyperbolic geometric space to better retain the relative distance and connectivity between the grass growth status features at each local position, reveal the hidden hierarchical structure between the features, and thus improve the feature expression capability.

[0046] More specifically, the step S322 is expressed by the formula:

[0047]

[0048] in,‖·‖ 2 represents the square of the norm of the eigenvector, represents the logarithmic function with base 2, s i Represents the fine-grained semantic significance measurement coefficient of the i-th grass growth state.

[0049] That is, in the hyperbolic space, by defining a metric function, the semantic significance of the grass growth status features at each pixel granularity is quantitatively evaluated, and by arranging the obtained semantic significance metric coefficients into a matrix form according to the corresponding pixel spatial positions, the fine-grained semantic significance distribution matrix of the grass growth status is obtained, thereby revealing the overall semantic significance distribution pattern of the multi-scale semantic encoding feature map of the grass growth status, intuitively showing the differences in the importance of the grass growth status features at different pixel positions, and providing an important basis for subsequent feature selection.

[0050] Specifically, in a specific example of the present application, the step S33 includes: inputting the fine-grained semantic saliency distribution matrix of the forage growth state into a feature explicit hierarchical precipitation module based on a multi-layer mask function to obtain a fine-grained semantic saliency distribution weight matrix of the forage growth state, which is expressed as:

[0051]

[0052] Where mask(·) represents a multi-layer mask function, θ represents the mask threshold, S represents the fine-grained semantic saliency distribution weight matrix of forage growth status, and s i ' represents the weight value of the ith position in the fine-grained semantic significance distribution weight matrix of the forage growth status.

[0053] That is, a multi-layer mask function is used to perform layered precipitation on the fine-grained semantic significance distribution matrix of the forage growth status, and different masks are applied to features of different significance levels to achieve selective amplification and suppression of features at different pixel positions in the multi-scale semantic encoding feature map of the forage growth status, thereby generating a corresponding weight matrix.

[0054] In a specific example of the present application, the step S33 further includes: calculating the fine-grained semantic significance distribution weight matrix of the forage growth state and the multi-scale semantic coding feature map of the forage growth state, multiplying the feature matrices along the channel dimension by position points to obtain the multi-scale semantic enhancement coding feature map of the forage growth state, which is expressed by the formula:

[0055] F f =F i ⊙S

[0056] Among them, ⊙ represents the point product by position, F f A multi-scale semantically enhanced coding feature map representing the growth status of the forage grass.

[0057] That is, by performing positional point multiplication operations on each feature matrix along the channel dimension of the original multi-scale semantic encoding feature map of the grass growth status, weighted adjustment of the original features is achieved, thereby significantly enhancing the differences in grass growth status features at different pixel positions in the image, helping to identify more subtle grass growth states, and providing more accurate feature support for subsequent grass growth status identification and classification.

[0058] In the above-mentioned forage planting humidity control method, the step S4 identifies the forage growth state label based on the multi-scale semantic enhancement coding feature map of the forage growth state. In a specific example of the present application, the step S4 includes: inputting the multi-scale semantic enhancement coding feature map of the forage growth state into a forage growth state identifier based on a classifier to obtain the forage growth state label. Among them, the classifier adopts a multi-layer perceptron structure, and through the layer-by-layer processing and feature abstraction of the fully connected layer, it can extract high-level features related to the forage growth state from the multi-scale semantic enhancement coding feature map of the forage growth state, and through the softmax function of the output layer, the extracted high-level features are mapped into a probability distribution for representing the forage growth state belonging to each predefined category, thereby determining the final output result, that is, the forage growth state label, such as "germination period", "tillering period", "jointing period", "maturity period", etc.

[0059] Specifically, the multi-scale semantic enhancement coding feature map of the grass growth state is input into a grass growth state identifier based on a classifier to obtain the grass growth state label, including: using the fully connected layer of the grass growth state identifier to fully connect the multi-scale semantic enhancement coding feature map of the grass growth state to obtain a multi-scale semantic fully connected coding feature vector of the grass growth state; inputting the multi-scale semantic fully connected coding feature vector of the grass growth state into the Softmax classification function of the grass growth state identifier to obtain the probability value of the multi-scale semantic fully connected coding feature vector of the grass growth state belonging to each predefined category, wherein the predefined category is the grass growth state label; and determining the predefined category corresponding to the largest one among the probability values ​​as the grass growth state label.

[0060] In the above-mentioned forage planting humidity control method, the step S5 determines the irrigation strategy based on the forage growth state label and the forage growth knowledge map. Figure 5 FIG. 5 is a flow chart of sub-step S5 in the forage planting humidity control method according to an embodiment of the present application. Figure 5As shown, the step S5 includes: S51, inputting the grass growth status label into the grass growth knowledge graph to obtain a recommended soil moisture value; S52, receiving a real-time soil moisture value collected by a humidity sensor; S53, determining an irrigation strategy based on a comparison between the real-time soil moisture value and the recommended soil moisture value.

[0061] Specifically, in step S51, the grass growth status label is input into the grass growth knowledge graph to obtain the recommended soil moisture value. It should be understood that the grass at each growth stage has different water requirements. For example, more water may be required to support the growth of new branches during the tillering period, while the amount of irrigation needs to be reduced during the dormant period. Here, the grass growth knowledge graph integrates rich agricultural science data and includes specific requirements for soil moisture at different growth stages of grass. Based on the grass growth knowledge graph, the soil moisture value that is most suitable for grass growth in this state can be queried according to the input growth status label, thereby realizing data-driven irrigation decisions.

[0062] Specifically, the step S52 receives the real-time soil moisture value collected by the humidity sensor. In order to ensure that the humidity sensor can stably and accurately collect and transmit soil moisture data, each link from sensor selection and installation to data reception and processing needs to be carefully designed and implemented. The specific implementation method of this process will be described in detail below:

[0063] First of all, choosing the right humidity sensor is the first step to ensure data accuracy and reliability. There are many types of humidity sensors on the market, including contact and non-contact types. Contact humidity sensors are directly inserted into the soil to measure moisture content, while non-contact sensors estimate soil moisture through electromagnetic wave or microwave reflection. For forage cultivation, contact sensors are more commonly used because they can directly reflect the moisture conditions inside the soil. When choosing a specific model, factors such as accuracy, response speed, durability, communication interface, and cost-effectiveness need to be considered. High-precision sensors can provide more accurate data, which helps to fine-tune irrigation strategies; fast-responding sensors can update data in time when the environment changes; considering the environmental challenges of outdoor use, sensors should have good waterproof and dustproof performance (such as IP67 and above) and withstand extreme temperature and humidity conditions; sensors that support Wi-Fi, 4G / 5G, LoRaWAN or other wireless communication protocols are convenient for remote monitoring and data transmission; choose cost-effective products according to budget and needs to meet functional requirements and control costs.

[0064] The installation location of the sensor directly affects the representativeness of the data. In order to obtain the most accurate soil moisture information, the sensor should be placed at a location that represents the average moisture of the entire field. It is usually recommended to bury the sensor about 10-20 cm underground. This depth can reflect the moisture conditions near the root system without being affected by surface evaporation. For large areas of farmland, it may be necessary to arrange multiple sensors reasonably and cover the entire monitoring area with reasonable spacing to ensure the comprehensiveness and representativeness of the data. In addition, the distance between sensors should be adjusted according to factors such as crop type and soil texture to avoid misjudgment due to local differences. Choosing an appropriate installation location and layout method can ensure that the collected data is highly representative, thereby providing a reliable basis for subsequent analysis and decision-making.

[0065] In order to achieve real-time transmission of humidity sensor data, it is crucial to build an efficient and stable communication network. According to the actual application scenario, you can choose wireless LAN (Wi-Fi), cellular network (4G / 5G), low-power wide area network (LoRaWAN) or Zigbee and Bluetooth Low Energy (BLE) and other communication solutions. Regardless of the communication method used, the security and reliability of data transmission should be ensured, encryption measures should be taken to protect data privacy, and redundancy mechanisms should be set up to prevent data loss. An efficient communication network not only ensures the timely transmission of data, but also provides a solid foundation for subsequent data processing and application.

[0066] Once the moisture sensor successfully collects and transmits soil moisture data, the next step is how to effectively receive and process this information. This step involves the selection of hardware platforms and the development of software systems. Select appropriate edge computing devices as data receiving terminals, such as Raspberry Pi, Arduino or other embedded systems, which can easily connect various types of sensors and run lightweight operating systems such as Linux for data processing. If you want to further analyze and store large amounts of historical data, you can consider using products from cloud computing service providers, such as Alibaba Cloud, AWS, Microsoft Azure, etc. The cloud platform provides powerful computing power and flexible expansion options, allowing users to dynamically adjust resource allocation according to actual needs. The received data is often a raw binary stream or a message body encapsulated by a specific protocol, so it is necessary to write a corresponding parsing program to convert it into a standard format that is easy to understand and process, such as JSON, CSV, etc. Establish a complete set of anomaly detection algorithms, so that when sensor failures, abnormal data fluctuations, etc. are found, alarms can be issued in time to notify relevant personnel to handle them. Ensure that all data collected by sensors have accurate timestamp information, which is very important for subsequent data analysis, especially when comparing multiple points or trend predictions. Effective data reception and preliminary processing provide accurate and reliable information support for subsequent applications and decisions.

[0067] Even if you select a high-quality humidity sensor and establish a reliable communication network, some error accumulation problems are inevitable during long-term operation. It is very necessary to calibrate the sensor on site regularly. You can check whether the sensor readings are accurate by comparing them with known standard samples, and adjust the parameters accordingly to ensure the consistency and reliability of the data. At the same time, you can also combine other auxiliary means such as rainfall records provided by weather stations to cross-verify the accuracy of humidity sensor data. The process of data verification and calibration ensures the quality and reliability of data in long-term operation, and provides a guarantee for the continuous optimization of irrigation management and agricultural production.

[0068] In order to make it easier for relevant technicians to understand the data collected by humidity sensors, it is very important to develop an intuitive and easy-to-use user interface. This can be achieved through web applications or mobile applications, providing functions such as real-time viewing, historical query, and chart display. Good user experience design not only improves work efficiency, but also promotes information transparency in the decision-making process. For example, users can intuitively see the trend of soil moisture changes in different time periods through a graphical interface, which helps to better understand the growth status of forage grass and make more scientific and reasonable irrigation decisions. By integrating sensor data acquisition, communication network construction, data reception and processing, verification and calibration, and user interface development, a complete solution is formed to ensure that the real-time soil moisture values ​​collected by humidity sensors can be effectively applied to precision agriculture practices, promote the development of agriculture towards intelligence, and achieve maximum resource utilization and sustainable development of the ecological environment.

[0069] Specifically, the step S53 determines the irrigation strategy based on the comparison between the real-time soil moisture value and the recommended soil moisture value. For example, a reasonable threshold range is set according to the recommended soil moisture value fluctuating up and down by a certain proportion (for example, ±5%). When the real-time soil moisture value is within this range, it is considered that the current moisture condition is moderate and no irrigation measures need to be taken immediately; if it is lower than the lower limit, it indicates that the soil is too dry and irrigation should be started; if it is higher than the upper limit, it indicates that there may be a risk of over-irrigation, and it is recommended to reduce the irrigation amount or suspend irrigation. The irrigation control system can automatically turn on or off water pumps, valves and other equipment according to preset rules to ensure that the operation is carried out according to the established plan. At the same time, in order to ensure the effective implementation of the irrigation strategy, it is necessary to regularly monitor the changes in soil moisture after irrigation and compare them with the expected results. If it is found that the actual effect does not meet expectations, the irrigation strategy should be adjusted in time, the parameter settings should be optimized, and the forage growth environment should always be in the best state. The irrigation records and effect evaluation reports are displayed through the user interface, which is convenient for managers to understand the situation at any time and make further adjustments.

[0070] In summary, the forage planting humidity control method based on the embodiment of the present application is explained, which uses deep learning-based image processing technology to process the forage growth state image to extract the multi-scale semantic features of the forage growth state, and enhances the feature significance of the forage growth state features at the pixel granularity to improve the feature distinction, capture more detailed forage growth state information, and thus intelligently identify the growth state label of the forage. On this basis, it is further combined with the forage growth knowledge graph to provide the forage with the most suitable soil moisture and irrigation strategy for its current growth state. In this way, it can ensure that the forage obtains appropriate water supply at different growth stages.

[0071] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to be implemented by adopting the above specific details.

Claims

1. A method for controlling humidity in forage planting, characterized in that: include: Acquire the grass growth status image collected by the camera; Extracting grass growth state features of the grass growth state image to obtain a grass growth state multi-scale semantic coding feature map; Performing feature saliency differentiation and enhancement based on pixel granularity on the multi-scale semantic coding feature map of the forage growth state to obtain a multi-scale semantic enhancement coding feature map of the forage growth state; Based on the multi-scale semantic enhancement coding feature map of the forage growth state, identifying the forage growth state label; Based on the forage growth status label and the forage growth knowledge graph, an irrigation strategy is determined.

2. The forage grass planting humidity control method according to claim 1, characterized in that: Extracting the grass growth state features of the grass growth state image to obtain a grass growth state multi-scale semantic coding feature map, including: The grass growth state image is input into a grass growth state feature extractor based on the FPT model to obtain a multi-scale semantic coding feature map of the grass growth state.

3. The forage grass planting humidity control method according to claim 2, characterized in that: The multi-scale semantic coding feature map of the forage growth state is subjected to feature saliency differentiation enhancement based on pixel granularity to obtain a multi-scale semantic enhancement coding feature map of the forage growth state, including: Performing pixel granularity feature decoupling on the multi-scale semantic coding feature map of the forage growth state to obtain a set of pixel granularity local decoupling semantic feature vectors of the forage growth state; Extracting the semantic significance distribution pattern of the set of pixel-granular local decoupled semantic feature vectors of the forage growth state to obtain a fine-grained semantic significance distribution matrix of the forage growth state; Based on the fine-grained semantic saliency distribution matrix of the forage growth state, the multi-scale semantic coding feature map of the forage growth state is subjected to feature explicit hierarchical modulation to obtain the multi-scale semantic enhanced coding feature map of the forage growth state.

4. The forage grass planting humidity control method according to claim 3, characterized in that: Extracting the semantic significance distribution pattern of the set of pixel-granular local decoupled semantic feature vectors of the forage growth state to obtain a fine-grained semantic significance distribution matrix of the forage growth state includes: Inputting each of the grass growth state pixel granularity local decoupling semantic feature vectors in the set of the grass growth state pixel granularity local decoupling semantic feature vectors into a hyperbolic space mapper to obtain a set of hyperbolic space modulated grass growth state pixel granularity local decoupling semantic feature vectors; The fine-grained semantic saliency measurement coefficient of each hyperbolic space modulated grass growth state pixel granularity local decoupling semantic feature vector in the set of the hyperbolic space modulated grass growth state pixel granularity local decoupling semantic feature vector is calculated to obtain the grass growth state fine-grained semantic saliency distribution matrix formed by arranging multiple grass growth state fine-grained semantic saliency measurement coefficients.

5. The forage grass planting humidity control method according to claim 4, characterized in that: The hyperbolic space mapper is used to perform continuous linear transformation on the forage growth state pixel granularity local decoupling semantic feature vector using a first weight parameter matrix and a second weight parameter matrix to obtain the hyperbolic space modulated forage growth state pixel granularity local decoupling semantic feature vector.

6. The forage grass planting humidity control method according to claim 5, characterized in that: Based on the fine-grained semantic saliency distribution matrix of the forage growth state, the multi-scale semantic coding feature map of the forage growth state is subjected to feature explicit hierarchical modulation to obtain the multi-scale semantic enhanced coding feature map of the forage growth state, including: Inputting the fine-grained semantic saliency distribution matrix of the forage growth state into a feature explicit hierarchical precipitation module based on a multi-layer mask function to obtain a fine-grained semantic saliency distribution weight matrix of the forage growth state; The fine-grained semantic significance distribution weight matrix of the forage growth state is calculated and multiplied by each feature matrix along the channel dimension of the multi-scale semantic coding feature map of the forage growth state at each position point to obtain the multi-scale semantic enhancement coding feature map of the forage growth state.

7. The forage grass planting humidity control method according to claim 6, characterized in that: Based on the multi-scale semantic enhancement coding feature map of the forage growth state, identifying the forage growth state label includes: The multi-scale semantic enhancement coding feature map of the grass growth state is input into a grass growth state identifier based on a classifier to obtain the grass growth state label.

8. The forage grass planting humidity control method according to claim 7, characterized in that: Inputting the multi-scale semantic enhancement coding feature map of the forage growth state into a forage growth state identifier based on a classifier to obtain the forage growth state label includes: Using the fully connected layer of the forage growth state identifier to perform fully connected encoding on the forage growth state multi-scale semantic enhancement encoding feature map to obtain a forage growth state multi-scale semantic fully connected encoding feature vector; Inputting the multi-scale semantic fully connected encoding feature vector of the forage growth state into the Softmax classification function of the forage growth state identifier to obtain the probability value of the multi-scale semantic fully connected encoding feature vector of the forage growth state belonging to each predefined category, wherein the predefined category is a forage growth state label; The predefined category corresponding to the maximum probability value is determined as the forage growth status label.

9. The forage grass planting humidity control method according to claim 8, characterized in that: Based on the grass growth status label and the grass growth knowledge graph, an irrigation strategy is determined, including: Inputting the forage growth status label into the forage growth knowledge graph to obtain a recommended soil moisture value; Receive real-time soil moisture values ​​collected by a moisture sensor; Based on the comparison between the real-time soil moisture value and the recommended soil moisture value, an irrigation strategy is determined.

Images