Intelligent ecological planting method of morchella under forest based on environment monitoring

Abstract

The present application relates to the field of intelligent planting of traditional Chinese medicinal materials and agricultural Internet of Things, and discloses a method for intelligent ecological planting of Morinda officinalis under forest canopy based on environmental monitoring, comprising: S1: determining the target canopy density range based on the forest type, and continuously acquiring the environmental parameters and leaf images of Morinda officinalis growth; S2: implementing ecological improvement of soil under forest canopy and deployment of water regulation infrastructure; S3: identifying and locating the disease spot based on the leaf images, segmenting the disease spot area, extracting the image features and fusing them with the corresponding environmental parameters to construct a multi-source feature set; using the improved NHWOA algorithm to optimize the risk assessment model and output the disease and pest risk grade; S4: making decision based on the real-time environmental parameters and the disease and pest risk grade, generating the regulation instruction and driving the corresponding equipment to execute, and adjusting the decision according to the execution feedback to form a closed loop. The present application realizes real-time sensing and accurate assessment of the microenvironment and disease and pest risk of Morinda officinalis under forest canopy.

Description

Technical Field

[0001] This invention relates to the field of intelligent cultivation of Chinese medicinal herbs and the Internet of Things in agriculture, specifically to an intelligent ecological cultivation method for *Gnaphalium affine* under forest cover based on environmental monitoring. Background Technology

[0002] As a precious traditional Chinese medicine, *Gnaphalium affine* has demanding requirements for its growth under forest microenvironment (light, temperature, humidity, soil physicochemical properties, etc.). Furthermore, karst topography presents challenges such as susceptibility to drought and flooding, poor soil quality, and complex habitats. Traditional cultivation methods rely on manual experience for regulation, resulting in low accuracy and delayed response. Simultaneously, the occurrence of pests and diseases in *Gnaphalium affine* (such as soft rot and leaf spot) is highly correlated with environmental parameters. Existing pest and disease identification largely relies on single image features, and early warning models employ traditional optimization algorithms (such as WOA and PSO). These algorithms suffer from insufficient population diversity, premature convergence, and susceptibility to local optima, leading to low risk assessment accuracy and inaccurate threshold search, making it difficult to meet the needs of refined cultivation.

[0003] Furthermore, existing environmental monitoring networks for understory cultivation mostly collect data based on single parameters, resulting in low data integration. Moreover, control measures such as irrigation and canopy density lack coordination, failing to achieve integrated management of light, water, soil, and pests. These problems directly lead to low survival rates, unstable yields, and inconsistent quality in *Mammillaria pubescens* cultivation, severely hindering the development of large-scale ecological cultivation of *Mammillaria pubescens*. Summary of the Invention

[0004] The core objective of this invention is to address the problems of low survival rate, unstable yield, and inconsistent quality in existing *Gnaphalium affine* cultivation methods. It proposes an intelligent ecological cultivation method for *Gnaphalium affine* under forest cover based on environmental monitoring. This method integrates multi-source sensing, intelligent algorithm optimization, and closed-loop control technologies for ecological cultivation and pest and disease early warning and management, making it particularly suitable for special habitats such as karst landforms.

[0005] To achieve the above objectives, the following technical solution is adopted: A method for intelligent ecological cultivation of *Gnaphalium affine* under forest cover based on environmental monitoring, characterized by the following steps: S1: Determining the appropriate target canopy density range based on the forest stand type of the target planting area, and deploying a monitoring network to continuously acquire environmental parameters and leaf images of *Gnaphalium affine* growth; S2: Implementing the deployment of understory soil ecological improvement and water regulation infrastructure, including: standardizing land preparation in the planting area, applying fully decomposed organic fertilizer, and introducing earthworms for ecological improvement; constructing a layered drainage network composed of main ditches, branch ditches, and furrows, and simultaneously deploying an intelligent drip irrigation network with pressure sensing and filtration functions; S3: Based on the leaf... The image is used to identify pests and diseases and locate lesions; the lesion area is segmented, and its morphological, color and texture features are extracted and fused with the corresponding spatiotemporal environmental parameters to construct a multi-source feature set; the risk assessment model constructed based on the multi-source feature set is optimized using the improved microhabitat hybrid heuristic whale optimization algorithm (NHWOA), and the pest and disease risk level is output based on the optimized risk assessment model; S4: the real-time environmental parameters of the monitoring network and the pest and disease risk level are used to make a decision judgment, and the control instructions that integrate irrigation, canopy closure adjustment and plant protection measures are generated based on the decision logic, and the corresponding equipment is driven to execute. Then, the decision is adjusted according to the execution feedback to form a closed loop.

[0006] Furthermore, in step S1, the forest stand type includes coniferous forest, broad-leaved forest and mixed forest, and the corresponding target canopy closure ranges are 0.6-0.7, 0.5-0.6 and 0.55-0.65, respectively.

[0007] Further, in step S3, a multi-source feature set is constructed, including: using a YOLOv8 target detection model to perform preliminary identification on the leaf image, outputting the category, confidence level, and bounding box coordinates of the lesions; locating the lesion region based on the bounding box coordinates, and using a Mask R-CNN semantic segmentation model to perform pixel-level segmentation on the lesion region to obtain an accurate binary lesion mask; calculating and extracting the morphological features, color features, and texture features of the lesions based on the binary lesion mask; wherein, the morphological features include at least the lesion area, perimeter, and roundness; the color features include at least the mean values ​​of each channel of the lesion region in the RGB color space and the peak value of the histogram in the HSV color space; the texture features include at least the energy and entropy calculated based on the gray-level co-occurrence matrix; and fusing the extracted morphological features, color features, and texture features with four environmental parameters acquired simultaneously when acquiring the leaf image: air temperature, relative humidity, soil moisture content, and forest light intensity, to jointly constitute a 15-dimensional multi-source feature vector for risk assessment.

[0008] Further, in step S3, the improved Hybrid Heuristic Whale Optimization Algorithm (NHWOA) optimizes the risk assessment model, including: A1. Initialization: setting algorithm parameters, including population size and maximum number of iterations, and dividing the population into multiple independent niches, each niche including several individuals; A2. Iterative optimization: each niche independently iterates and updates based on its local optimum, with update strategies including prey encirclement, random search, and spiral attack; A3. Diversity maintenance: randomly redistributing individuals in all niches when a preset redistribution period is reached; A4. Perturbation and escape: applying random perturbations based on Levy flight to the individual positions during iteration; A5. Parameter adaptation: using a first probability parameter value to focus on global exploration in the early stage of iteration, and a second probability parameter value to focus on local development in the later stage of iteration; repeating steps A2 to A5 until the maximum number of iterations or convergence conditions are met, and outputting the optimal feature weight set and risk threshold.

[0009] Furthermore, the goal of optimizing the risk assessment model is to maximize the accuracy of pest and disease risk assessment. This is specifically achieved by synchronously and iteratively optimizing the weights of each feature in the risk assessment model. and risk assessment threshold To achieve this, the formula for calculating the integrated pest and disease risk value R is:

[0010] ;

[0011] in, For the first in the multi-source feature set One characteristic, The corresponding feature weights obtained by optimizing the NHWOA algorithm, and n is the total number of features; when the comprehensive risk value R of pests and diseases is greater than or equal to the preset risk threshold. At that time, an early warning is triggered.

[0012] Furthermore, in step S4, the decision logic is configured to: generate and execute a first type of control instruction when only a single environmental parameter deviates from the preset range; and generate a second type of comprehensive control instruction that integrates multiple control measures and drives the corresponding equipment to execute when multiple environmental parameters are abnormally cross-linked or the pest and disease risk level exceeds the preset threshold.

[0013] Furthermore, when only a single environmental parameter deviates from the preset range, a first type of control instruction is generated and executed, including: rapid closed-loop control based on a single parameter threshold, independently triggering and executing precision irrigation or forest canopy closure adjustment based on real-time soil moisture content and light intensity data; wherein, the condition for triggering the precision irrigation is that the percentage of soil moisture content to field capacity is less than 80%; the precision irrigation amount Q is calculated based on the following formula:

[0014] ;

[0015] Where Q is the irrigation amount, Where is the area of ​​the zone, and h is the depth of the soil wetting layer. It refers to the soil field water holding capacity. This represents the current soil moisture content as a percentage of field capacity. This is the density of water.

[0016] Furthermore, the adjustment of forest canopy density includes: when the understory light intensity... At that time, selective pruning is carried out on the tree layer; when the light intensity under the forest canopy is high... When doing so, prioritize setting up shade nets or preserving the lower branches of trees.

[0017] Furthermore, when generating the second type of comprehensive control command, the following decision model is invoked to perform multi-objective trade-offs: ;in, This is the comprehensive decision value when multiple parameters are abnormal; For the baseline control strategy targeting the j-th abnormal parameter, These are the corresponding weighting coefficients.

[0018] Furthermore, step S4 includes:

[0019] The generated control commands are sent to the corresponding drip irrigation system, drainage system or forest stand control device for execution. After execution, a status confirmation signal is returned to the decision-making unit. The decision-making unit updates the system status based on the returned signal and determines whether the updated environmental parameters have returned to the appropriate range. If they have not returned, the decision-making process is retried to generate new control commands, thereby realizing closed-loop control of monitoring-decision-execution-feedback.

[0020] Compared with the prior art, the present invention achieves the following beneficial effects:

[0021] 1. Significant algorithm optimization results: The NHWOA algorithm effectively solves the problems of insufficient population diversity and premature convergence of traditional optimization algorithms such as WOA, PSO, and GA by dividing niches and assigning individual weights. It improves the accuracy of multi-source feature weight optimization and risk threshold search, and increases the accuracy of pest and disease risk assessment to over 92.8%, providing core technical support for accurate early warning.

[0022] 2. Multi-source fusion improves assessment reliability: By integrating image features such as lesion morphology, color, and texture with environmental data such as air temperature and humidity and soil moisture content, a 15-dimensional multi-source feature set is constructed, achieving comprehensive coverage of "lesion characteristics + growth environment characteristics", which significantly reduces the false positive rate of risk assessment.

[0023] 3. Synergistic regulation to adapt to special habitats: In response to the characteristics of karst landforms, a water regulation system of "three-level drainage + intelligent drip irrigation" and a dynamic control mechanism for canopy density were constructed. Through the weighted decision model optimized by NHWOA, synergistic regulation under multiple parameter anomalies was achieved, and the survival rate of Maocigu mushroom cultivation was increased to over 90%, effectively solving the problems of "easy to drought and flood" and "uneven light".

[0024] 4. Efficient and convenient closed-loop management throughout the entire process: It realizes fully automated closed-loop management from environmental monitoring, data preprocessing, risk warning to control execution, with a total response time of ≤30 seconds, which greatly reduces the cost of manual intervention and supports tiered configuration to adapt to the actual needs of different planting scales.

[0025] In summary, the embodiments of the present invention construct a complete technical system of multi-parameter monitoring, multi-source feature fusion, intelligent early warning, and collaborative regulation, so as to realize real-time perception and accurate assessment of the microenvironment (light, temperature, humidity, and soil parameters) and pest and disease risks under the Maocigu forest.

[0026] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0027] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the invention. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0028] Figure 1 This is a flowchart illustrating the intelligent ecological cultivation method for *Gnaphalium affine* under forest cover based on environmental monitoring, according to an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of the overall process of the intelligent ecological cultivation method for *Gnaphalium affine* under forest based on environmental monitoring, according to an embodiment of the present invention.

[0030] Figure 3 This is a structural diagram of the multi-source feature fusion and intelligent early warning model in an embodiment of the present invention;

[0031] Figure 4 These are screenshots of a mobile terminal APP according to an embodiment of the present invention;

[0032] Figure 5 This is a comparison chart of the recognition accuracy of the algorithm of this invention and the traditional algorithm. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0034] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0035] Figure 1 This is a flowchart illustrating the intelligent ecological cultivation method for *Gnaphalium affine* under forest cover based on environmental monitoring, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the overall process of the intelligent ecological cultivation method for *Gnaphalium affine* under forest cover based on environmental monitoring, according to an embodiment of the present invention. Figure 1 and Figure 2 As shown, the intelligent ecological cultivation method for *Gnaphalium affine* under forest cover based on environmental monitoring includes:

[0036] S1: Determine the appropriate target canopy closure range based on the forest stand type of the target planting area, and set up a monitoring network to continuously acquire environmental parameters and leaf images of the hairy taro growth.

[0037] Step S1 is used to initialize planting environment parameters and construct a multi-parameter environmental monitoring network, including the following steps:

[0038] S1.1: Initialization of planting environment parameters

[0039] S1.1.1: Stand Adaptation Parameter Calibration

[0040] Based on existing forestry resource databases (such as the National Forestry and Grassland Administration's vegetation monitoring database) and research literature on the understory cultivation of *Gnaphalium affine*, we extracted understory microenvironmental characteristic data for coniferous forests, broad-leaved forests, and mixed forests to determine the target canopy closure ranges for different forest stand types: coniferous forests (Dense coniferous canopy requires some light penetration) Broadleaf forest (Broad-leaved leaves are large and have good light transmission uniformity), mixed forest (Taking into account the light transmission characteristics of both coniferous and broad-leaved leaves).

[0041] S1.1.2: Core Environmental Parameter Baseline Settings

[0042] Based on the growth habits of *Gnaphalium affine* and the characteristics of karst soil, the suitable ranges for key environmental parameters were determined (the following are the baselines for the main vegetative growth period, which can be slightly adjusted according to phenological stages): Air temperature relative humidity of air Soil temperature Soil moisture content (as a percentage of field capacity) (Suitable for sandy loam soil, field water holding capacity) Range of values (volume water content), soil pH value (Suitable for acidic soils), forest understory light intensity Avoid strong light burns and weak light etiolation. Soil field water holding capacity (volume water content) is the core benchmark parameter for calculating the proportion of soil moisture content.

[0043] S1.1.3: Baseline Data Acquisition and Application

[0044] In the proposed planting area Grid layout At each sampling point, a portable high-precision monitoring device (soil pH meter accuracy) was used. Light sensor accuracy Collect initial values ​​of the above parameters and construct a benchmark database. This database is primarily used to assess whether the initial conditions of the planting area are suitable (if the initial soil pH value deviates from the target range). Organic matter content (This requires prior optimization) and serves as a historical reference for environmental changes during the growth cycle; during dynamic regulation, the "suitable range" is used directly as the decision-making basis, rather than a fixed initial value. The data collection period is three consecutive days, once each in the morning, noon, and evening.

[0045] The initial environmental baseline database for the understory cultivation area of ​​*Gnaphalium affine* is used to assess the suitability of the initial conditions of the cultivation area and to serve as a historical reference for environmental changes during the growth cycle. The initial air temperature in the planting area is one of the core parameters of the baseline database, and the unit is... . The initial relative humidity of the planting area is one of the core parameters of the baseline database, and the unit is... . The initial soil temperature in the planting area is one of the core parameters of the baseline database, and the unit is... . The initial soil moisture content of the planting area (as a percentage of field capacity) is one of the core parameters of the baseline database. The initial soil pH value in the planting area is one of the core parameters of the benchmark database, and its value range must meet the following requirements. To suit the growth of Sagittaria sagittifolia. Initial forest understory light intensity in the planting area is one of the core parameters of the baseline database, and its unit is lux.

[0046] S1.2: Construction of a Multi-Parameter Environmental Monitoring Network

[0047] S1.2.1: Deployment of Monitoring Nodes

[0048] according to Multiple monitoring nodes are deployed in the understory planting area to monitor density. Each node integrates multi-parameter sensors and an image acquisition module: air temperature and humidity sensor (measurement range...) , precision Soil temperature and humidity sensor (measuring range) precision , Soil pH sensor (measuring range) (Accuracy ±0.1), Light sensor (measurement range) 20,000 lux, accuracy ±10 lux), high-definition camera (resolution) With a frame rate of 30fps and support for autofocus, its core function is to acquire images of *Sagittaria sagittifolia* leaves, providing a data source for intelligent pest and disease identification in step six. All electronic devices meet the IP67 industrial protection standard, making them suitable for humid and dusty environments under forest canopies.

[0049] S1.2.2: Data Transmission and Preprocessing

[0050] Monitoring data is transmitted via LoRa wireless communication technology (in open areas under forest cover with good line-of-sight conditions, the transmission distance is [not specified]). Transmission distance in densely vegetated areas under forest canopy Power consumption The data is transmitted to the data acquisition module (industrial-grade gateway, supporting edge computing) using a moving average filtering method. ( To reduce random noise interference, noise reduction is applied to the original data, and linear interpolation is used for missing data. The formula is as follows: Completed by using effective data from adjacent time points Missing data at any given time. Layered data update frequency: soil temperature and humidity, pH value every 10 minutes; air temperature and humidity, light intensity... Minutes per instance (automatically switching to 1 minute per instance in case of sudden weather changes), ultimately forming a standardized real-time dataset. ( (The image data of the blade at time t) ensures the timeliness of monitoring.

[0051] Environmental monitoring data after being processed by moving average filtering is used to reduce random noise interference in the original data. : The window size for the moving average filter, which is set to 5, meaning that the average is calculated using 5 consecutive raw data points. Completed using linear interpolation Real-time environmental monitoring data is used to address the issue of missing data. : Raw environmental monitoring data from the moment preceding the current moment. : Raw environmental monitoring data one moment after the previous moment. Standardized real-time datasets provide real-time environmental data support for dynamic regulation. : Real-time air temperature readings at any given time, in units of . : Real-time relative humidity readings at any given time, in units of . Real-time soil temperature monitoring value at any given time, in units of . Real-time soil moisture content (as a percentage of field capacity). : Real-time monitoring value of soil pH at any given moment. : Real-time monitoring value of forest undergrowth intensity at any given time, in lux. : The constantly collected images of Maocigu leaves provide a data source for intelligent identification of diseases and pests.

[0052] S2: Implement the deployment of forest understory soil ecological improvement and water regulation infrastructure, including: standardizing the land preparation of the planting area, applying fully decomposed organic fertilizer and introducing earthworms for ecological improvement; constructing a layered drainage network consisting of main ditches, branch ditches and furrows, and simultaneously deploying an intelligent drip irrigation network with pressure sensing and filtration functions.

[0053] Step S2 is used to implement forest understory soil ecological improvement treatment and infrastructure deployment, including the following steps:

[0054] S2.1: Understory ecological engineering and soil improvement treatment

[0055] S2.1.1: Standardized land preparation operations

[0056] Prepare the land according to planting specifications. First, remove stones, weeds, and dead leaves from the plot. Use a rotary tiller for deep plowing (15-20cm deep) to break up the compacted soil layer. Then, mark rows with a spacing of 20cm x 30cm and dig furrows (5-8cm deep). Spread fully decomposed organic fertilizer (mainly sheep manure and leaf compost, at a rate of 2000-3000kg / mu) at the bottom of the furrows and mix it evenly with the soil. Level the plot afterward. It is emphasized that the organic fertilizer must be fully decomposed (decomposition time ≥6 months, compost temperature ≥55℃ and maintained for more than 3 days) to avoid introducing pathogens, causing root burn, or attracting underground pests.

[0057] S2.1.2: Ecological Improvement Measures

[0058] Soil biological improvement was carried out by introducing Eisenia fetida. The optimal release time was spring and autumn (soil temperature 15-25℃, moisture content 60-70%, suitable for earthworm survival), with a release density of 1500-2000 earthworms per acre. After release, a 5cm layer of well-rotted straw was used as a cover (to maintain soil moisture and provide a food source for the earthworms). Earthworm activity improved soil aggregate structure, increased aeration and water retention capacity, and promoted the decomposition and conversion of organic fertilizer, raising the soil organic matter content to over 1.5%. Soil physicochemical parameters were monitored regularly (every 15 days). If the pH value deviated from the 5.5-6.5 range, it was slightly adjusted by applying leaf mold (to adjust acidity) or a small amount of lime (to adjust alkalinity) to ensure that soil conditions were suitable for the growth requirements of *Sagittaria sagittifolia*.

[0059] S2.1.3: Verification of Improvement Effect

[0060] After land preparation and ecological improvement are completed, soil samples are collected from 10 representative sampling points to test soil moisture content, pH value, organic matter content and permeability. The following conditions must be met: soil bulk density ≤1.3g / cm³, porosity ≥45%, and organic matter content ≥1.5%. Otherwise, additional plowing or adjustment of organic fertilizer application is required.

[0061] S2.2: Drainage system construction and water regulation infrastructure deployment

[0062] S2.2.1: Layered drainage system

[0063] To address the drought- and flood-prone characteristics of karst topography, a three-tiered drainage network of "main ditch - branch ditch - ridge ditch" was constructed along contour lines. The main ditch is 30cm deep and 20cm wide, with a 0.3% slope at the bottom (to ensure smooth drainage), and is lined with a 5cm thick layer of crushed stone to enhance drainage capacity. Branch ditches are 20cm deep and 15cm wide, spaced 10-15m apart, and vertically connected to the main ditch. Ridge ditches are 15cm deep and 10cm wide, laid along planting ridges, with one ditch every 3-4 ridges, connecting to the branch ditch. The ditch walls are compacted to prevent soil collapse and blockage.

[0064] S2.2.2: Deployment of Intelligent Drip Irrigation System

[0065] The drip irrigation network is laid after the final fine land preparation and before the cultivation of *Cymbidium goeringii* (a type of mushroom). It uses buried PE drip irrigation pipes (16mm diameter, working pressure 0.15~0.2MPa), with a dripper spacing of 50cm, a flow rate of 2~3L / h, and a burial depth of 5~10cm (to avoid aging from direct sunlight and damage to agricultural machinery). Clearly marked posts are erected in the pipe laying area to prevent damage during subsequent agricultural operations. The system is equipped with a variable frequency water pump (flow rate 0~5m³ / h), an electromagnetic control valve (response time ≤0.5s), a pressure sensor (measurement range 0~0.4MPa), and a Y-type filter (120-mesh filtration accuracy to handle high-hardness water in karst areas and prevent dripper clogging). It achieves data linkage with the environmental monitoring network via the Modbus protocol, supporting precise command-based single-point, area, or full-area water replenishment operations, and also has a manual emergency control function.

[0066] S2.2.3: Facility Commissioning and Maintenance

[0067] After deployment, conduct joint commissioning of the drainage and irrigation systems: Start the drip irrigation system and run it continuously for 30 minutes. On flat land, test the uniformity of water output from the drippers (coefficient of variation ≤10%; for slightly undulating terrain, coefficient of variation ≤15%). Simulate rainfall (5L / m²) and observe the drainage efficiency of the drainage system, ensuring no surface water accumulation within 30 minutes. Establish a regular maintenance mechanism: flush the filters every 15 days and check the dripper patency every 30 days to ensure stable infrastructure operation.

[0068] S3: Based on the leaf image, identify pests and diseases and locate lesions; segment the lesion area, extract its morphological, color and texture features and fuse them with the corresponding spatiotemporal environmental parameters to construct a multi-source feature set; use the improved microhabitat hybrid heuristic whale optimization algorithm (NHWOA) to optimize the risk assessment model constructed based on the multi-source feature set, and output the pest and disease risk level based on the optimized risk assessment model;

[0069] Step S3 is used to realize intelligent identification and early warning control of pests and diseases, such as Figure 3 The diagram shown is a structural diagram of the multi-source feature fusion and intelligent early warning model in an embodiment of the present invention. Step S3 includes the following steps:

[0070] S3.1: Data Source Acquisition and Balancing

[0071] Public database: Images of common diseases and pests of Mushroom sphagnum (soft rot, leaf spot, grub damage) were downloaded from authoritative platforms such as Plant Village and China Crop Disease and Pest Image Database, covering different disease levels (0-4), shooting angles (front, side, oblique angle) and lighting conditions (sunny day, cloudy day, forest diffused light), with a total of 2000+ images collected.

[0072] Field photography: Images were taken on artificially inoculated and naturally diseased plants in the experimental field of Maocigu cultivation to supplement the characteristics of pests and diseases at different growth stages (budding stage, vegetative growth stage, and bulb enlargement stage), with a total of 2,500+ images collected.

[0073] Simulated Samples: Leaf samples were created by artificially simulating lesions (such as water-soaked spots of soft rot and brown necrotic spots of leaf spot). The simulated lesions were ensured to be highly consistent with real lesions in texture, color, and edge features (visual consistency was assessed by plant pathology experts, with consistency ≥90%) to avoid model overfitting. A total of 500+ images were collected. The final dataset consisted of 5000+ labeled images (labels included pest / disease type, disease severity, and lesion location), divided into training, validation, and test sets in a 7:2:1 ratio.

[0074] S3.2 Constructing a Pest and Disease Identification Model

[0075] The YOLOv8 target detection model was used for lesion identification and classification. The model structure included: an input layer (image size 640×640, RGB three-channel), a C2f feature fusion module (8 modules to enhance feature extraction capabilities), a spatial pyramid pooling module (aggregating multi-scale features), and a detection head (classification head + regression head, outputting pest / disease type and lesion coordinates). Training steps: Data augmentation techniques such as random flipping, brightness adjustment, and cropping were used to expand the sample size. These data augmentation techniques are sample expansion methods used during model training to improve the model's generalization ability and adapt to leaf images under different lighting conditions and shooting angles in the forest. The optimizer used was AdamW (learning rate 0.001, weight decay 0.0005), the activation function was SiLU (to enhance gradient flow), and the loss function was CIoU loss (to optimize bounding box regression accuracy). Training was iterated for 100 epochs, and training was stopped when the accuracy on the validation set stabilized at ≥92%. The model input is a real-time image of a mushroom leaf, and the output is the type of disease or pest (soft rot / leaf spot / grub damage), the disease level (0~4), and the confidence level (≥0.7 for valid identification).

[0076] The YOLOv8 target detection model is used for the identification and classification of lesions in *Amanita muscaria* (a type of mushroom). It possesses efficient feature extraction and target localization capabilities, making it suitable for pest and disease detection scenarios using leaf images from forest understory. RGB three-channel: The color channel mode of the input image for the YOLOv8 model, corresponding to red, green, and blue channels, fully preserving the color information of the leaf image. C2f feature fusion module: The core feature extraction module of the YOLOv8 model, consisting of eight modules. It enhances the model's ability to extract lesion features through multi-scale feature fusion, improving the recognition accuracy of small lesions. Spatial pyramid pooling module (SPPF): The feature aggregation module in the YOLOv8 model, used to integrate lesion features at different scales, adapting to lesions of different sizes and shapes. Detection head: The output module of the YOLOv8 model, containing a classification head and a regression head, responsible for pest and disease type determination and lesion coordinate localization, respectively. Classification head: A submodule of the detection head, used to output the pest and disease type (soft rot / leaf spot / grub damage) and corresponding confidence score. Regression Head: A submodule of the detection head, used to output the coordinate information of lesions in the leaf image, providing a location basis for subsequent lesion region segmentation. AdamW: The optimizer of the YOLOv8 model, used to adjust model parameters to minimize the loss function. The learning rate is set to 0.001, and the weight decay is set to 0.0005, balancing model training efficiency and generalization ability. Learning Rate 0.001: The parameter update step size of the AdamW optimizer, controlling the adjustment magnitude of model parameters in each iteration to ensure stable training convergence. Weight Decay 0.0005: The regularization parameter of the AdamW optimizer, used to suppress model overfitting and improve the ability to identify unseen samples. SiLU: The activation function of the YOLOv8 model, improving gradient flow efficiency through nonlinear transformation, enhancing the model's ability to fit complex lesion features. CIoU Loss: The loss function of the YOLOv8 model, used to optimize the regression accuracy of lesion bounding boxes, comprehensively considering the overlap, distance, and aspect ratio of the bounding boxes to improve the accuracy of lesion localization.

[0077] S3.3: Multi-source feature fusion

[0078] To achieve accurate risk assessment of diseases and pests affecting *Gnaphalium affine*, this invention constructs a multi-source feature set integrating visual characteristics of lesions and environmental factors. The morphological characteristics of lesions (area, perimeter, circularity) reflect the stage and type of disease development; color characteristics (RGB mean, HSV peak) effectively distinguish tissue discoloration caused by pathogens; and texture characteristics (energy, entropy) characterize the degree of structural damage to lesion tissue. Simultaneously, environmental characteristics closely related to disease development (air temperature, humidity, soil moisture content, light intensity) are introduced because these parameters directly affect pathogen activity and plant resistance. Therefore, by fusing these four categories of 15 features, a comprehensive characterization of the lesion's state and pathogenic environmental conditions can be achieved, providing highly discriminative input for subsequent risk assessment models. The specific construction process is as follows:

[0079] Multi-source feature set The system is constructed using images of *Gnaphalium affine* leaves output by a pest and disease identification model as the core data source. All lesion-related features are extracted from precisely segmented lesion areas in the images and then spatiotemporally matched and fused with real-time environmental monitoring data and regional meteorological data to ensure the relevance and relevance of the features.

[0080] The 15-dimensional multi-source feature set for the risk assessment of diseases and pests of Mushroom styrax covers the characteristics of the lesions themselves and the characteristics of the growth environment, providing input data for the subsequent optimization of the NHUOA early warning model. : A single feature vector in the 15-dimensional feature set, corresponding to lesion morphology features (3-dimensional), color features (6-dimensional), texture features (2-dimensional), and environmental-meteorological fusion features (4-dimensional).

[0081] S3.3.1: Lesion area segmentation

[0082] Based on the bounding box coordinates of lesions output by the YOLOv8 model (precisely locating the approximate position of the lesions in the leaf image), the Mask R-CNN semantic segmentation algorithm is further used for pixel-level fine processing. The core objective is to completely separate the lesion region from the healthy leaf region, avoiding interference from healthy tissue in feature extraction. The specific operation process is as follows:

[0083] (1) Image preprocessing: First, the local leaf image after YOLOv8 positioning (centered on the lesion boundary box, extending outward by 20 pixels to avoid cropping the lesion edge) is standardized—the image size is uniformly adjusted to 512×512 pixels, and gamma correction is performed. =1.2) Balance the differences in image brightness caused by uneven lighting under the forest canopy, and then perform Gaussian blur (3×3 convolution kernel) to suppress high-frequency noise.

[0084] : Gamma correction factor, with a value of 1.2, is used to balance the differences in image brightness caused by uneven lighting under the forest canopy, thereby improving image contrast and detail clarity.

[0085] (2) Feature extraction and candidate region generation: Mask R-CNN uses ResNet50+FPN (Feature Pyramid Network) as the backbone network. First, multi-scale feature extraction is performed on the preprocessed image (generating four scale feature maps C2~C5). Then, candidate boxes are generated on the feature maps through the Region Proposal Network (RPN). The candidate boxes are filtered by combining the lesion bounding boxes output by YOLOv8 (IoU threshold ≥ 0.7). After filtering, only candidate regions highly related to lesions are retained, which improves the segmentation efficiency. C2~C5: The four scale feature maps output by the Mask R-CNN backbone network, which correspond to different levels of image features (C2 is shallow feature, C5 is deep feature), covering the details and semantic features of lesions.

[0086] (3) Mask generation and classification: For the selected candidate regions, the RoI Align layer maps the candidate regions of different scales onto a feature map of the same size, and performs two branch tasks in parallel: one is lesion / healthy tissue binary classification (outputting the confidence of the region as a lesion), and the other is generating a coarse mask of 14×14 pixels; then the coarse mask is upsampled to 512×512 pixels through a mask refinement network (3 layers of transposed convolution) to align with the original image size.

[0087] (4) Post-processing optimization: Morphological operations are performed on the generated refined mask. First, the small noise points (isolated areas with an area of ​​<5 pixels) are removed by erosion operation (3×3 rectangular structuring element). Then, the small holes in the lesion area are repaired by dilation operation (same structuring element). Finally, a binary mask image is obtained (pixel value of lesion area = 255, pixel value of healthy area = 0), ensuring the integrity and purity of the lesion area.

[0088] S3.3.2: Morphological Feature Extraction (3D: )

[0089] For the segmented lesion mask image, core morphological parameters are calculated based on digital image processing techniques:

[0090] (1) Area of ​​lesions : Count the total number of pixels within the lesion area, combined with the calibration ratio of image pixels to actual size (1 pixel corresponds to...). ), converted to actual area (unit: );

[0091] (2) Perimeter of the lesion The chain code tracing algorithm is used to extract the edge contour of the lesion, count the number of edge pixels, and convert them into the actual perimeter (unit: cm) according to the same calibration ratio.

[0092] (3) Circularity of lesions : Circularity through formula Calculate the range of values. The closer the value is to 1, the closer the lesion is to a circle. This value is used to distinguish the differences in lesion morphology between different diseases and pests (e.g., the water-soaked lesions of soft rot have a higher degree of roundness, while the angular lesions of leaf spot have a lower degree of roundness).

[0093] S3.3.3: Color Feature Extraction (6-dimensional: )

[0094] The segmented lesion region image was converted from the RGB color space to the HSV color space (to improve color stability under varying forest lighting conditions), and two types of color features were extracted respectively:

[0095] (1) RGB mean characteristics Calculate the area within the lesion. The average pixel values ​​of the three channels reflect the overall color intensity of the lesions (e.g., the average R channel value of brown spots in leaf spot disease is significantly higher than that of healthy leaves, and the average G channel value of water-soaked spots in soft rot disease is relatively high).

[0096] in, The RGB mean feature dimension, representing the color characteristics of lesions from multiple sources, corresponds to the mean pixel values ​​of the R (red), G (green), and B (blue) channels, respectively. R: The red channel in the RGB color space; its mean pixel value reflects the overall depth of red hues in the lesion area. For example, the mean R channel value of brown spots in leaf spot disease is significantly higher than that of healthy leaves. G: The green channel in the RGB color space; its mean pixel value reflects the overall depth of green hues in the lesion area. For example, the mean G channel value of water-soaked spots in soft rot disease is relatively high. B: The blue channel in the RGB color space; its mean pixel value reflects the overall depth of blue hues in the lesion area, serving as an important supplementary dimension to the color characteristics of lesions. The mean pixel value within the lesion area is calculated by taking the arithmetic mean of the R, G, and B channel values ​​of all pixels in the precisely segmented lesion area. This mean quantifies the overall color depth of the lesion and is a key color feature for distinguishing different types of pests and diseases.

[0097] (2) Peak characteristics of HSV histogram Histograms (bins) were constructed for the H (hue), S (saturation), and V (brightness) channels of the lesion area. Extracting the values ​​corresponding to the peak values ​​of the histogram to characterize the most significant color features of the lesions (e.g., the H channel peaks of fungal lesions are concentrated in...). In this area, bacterial diseases are concentrated (Interval).

[0098] in, : The peak value of the HSV histogram, representing the color characteristics of lesions from multiple sources, corresponds to the peak values ​​of the H (hue), S (saturation), and V (brightness) channels, respectively. bins=256: The number of groups in the HSV channel histograms, dividing the channel value range into 256 intervals to accurately capture subtle differences in lesion color. Histogram peak value: The value corresponding to the interval with the largest value in each HSV channel histogram, representing the most significant color feature of the lesion, and is a key basis for distinguishing different types of pests and diseases.

[0099] S3.3.4 Texture Feature Extraction (2D: )

[0100] The image of the lesion area was converted to grayscale (to an 8-bit grayscale image, pixel values). Texture features are extracted based on the Gray-Level Co-occurrence Matrix (GLCM), with the matrix parameter set to distance. Pixels, angle After calculation, the average of the four angles is taken as the final feature, where, Pixel: The distance parameter of the gray-level co-occurrence matrix, which is to statistically analyze the co-occurrence relationship of gray values ​​of adjacent pixels, and adapt to the fine features of lesion texture. : Angular parameters of the gray-level co-occurrence matrix, respectively statistically analyzed at the horizontal and vertical levels. oblique, vertical Gray-level co-occurrence features in four diagonal directions ensure comprehensive texture extraction. The average of the four angles is taken as the final energy and entropy values, eliminating the influence of angle bias on the features.

[0101] (1) Energy value The formula is: (These are grayscale co-occurrence matrix elements), reflecting the uniformity of the texture. The larger the value, the more regular the lesion texture (e.g., the texture energy value of bite spots caused by grub damage is lower, while the texture energy value of fungal lesions is higher).

[0102] in, The elements of the gray-level co-occurrence matrix represent gray-level values. pixels and distance ,angle grayscale value The probability of pixels appearing simultaneously.

[0103] (2) Entropy The formula is: It reflects the complexity of the texture. The larger the value, the more chaotic the texture of the lesion (such as the entropy value of lesions in the later stage of disease due to tissue necrosis, which is significantly higher than that of lesions in the early stage).

[0104] S3.3.5: Fusion of Environmental and Meteorological Data (4-dimensional: )

[0105] Extract real-time data that perfectly matches the time and spatial location of the leaf image acquisition: air temperature. relative humidity Soil moisture content Forest light intensity ; Supplement daily rainfall data in regional meteorological data (integrated into) Related calculations, or as a standalone calculation In a preferred embodiment of the present invention, the 15-dimensional adjustment is as follows: the HSV peak feature is simplified to 2-dimensional, while maintaining the final 15-dimensional structure, ensuring that the feature set covers "lesion characteristics + growth environment characteristics" to comprehensively support risk assessment. The final result is a 15-dimensional feature vector, where... For morphological characteristics, For color features, For texture features, This provides comprehensive and accurate input data for the subsequent optimization of the NHWOA early warning model by integrating environmental and meteorological characteristics.

[0106] The multi-source features are concentrated in the dimensions of environmental and meteorological characteristics, corresponding to air temperature, relative humidity, soil moisture content, and forest light intensity, respectively. : Spatiotemporal matching with blade image acquisition Real-time air temperature value, in units of This is one of the environmental characteristics in pest and disease risk assessment. : Spatiotemporal matching with blade image acquisition Real-time relative humidity value, in units of This affects the occurrence and spread rate of pests and diseases. : Spatiotemporal matching with blade image acquisition The soil moisture content at any given time (as a percentage of field capacity) is related to the development of lesions and plant resistance. : Spatiotemporal matching with blade image acquisition Real-time forest canopy light intensity, in lux; daily rainfall is integrated into this dimension for correlation calculation (or set separately). To adapt to the 15-dimensional feature structure, the HSV peak features are simplified to 2-dimensional, ultimately maintaining the 15-dimensional structure of the feature set.

[0107] S3.4: Constructing an NHW OA optimization early warning model

[0108] Traditional Whale Optimization (WOA) algorithms suffer from premature convergence (due to insufficient population diversity) and susceptibility to local optima, making them unsuitable for weight optimization and threshold search of multi-source features and failing to meet the requirements of multi-source feature optimization. Therefore, this invention proposes an improved Niching Hybrid Heuristic Whale Optimization Algorithm (NHWOA), an improved version of the traditional WOA algorithm that solves the problems of premature convergence and susceptibility to local optima, and is suitable for multi-source feature weight optimization and risk threshold search.

[0109] Specifically, the improved NHWOA (Hybrid Niche Heuristic Whale Optimization) algorithm in this embodiment of the invention optimizes the risk assessment model as follows: A1. Initialization: Set algorithm parameters, including population size and maximum number of iterations, and divide the population into multiple independent niches, each niche including several individuals; A2. Iterative Optimization: Each niche independently iterates and updates based on its local optimum, with update strategies including prey encirclement, random search, and spiral attack; A3. Diversity Maintenance: When a preset redistribution period is reached, individuals in all niches are randomly redistributed; A4. Perturbation and Escape: During the iteration process, random perturbations based on Levy flight are applied to the individual positions; A5. Parameter Adaptation: In the early stage of iteration, a first probability parameter value is used to focus on global exploration, and in the later stage of iteration, a second probability parameter value is used to focus on local development; Steps A2 to A5 are repeated until the maximum number of iterations or the convergence condition is met, and the optimal feature weight set and risk threshold are output. Optionally, as a specific embodiment, the specific process of the NHWOA algorithm is as follows:

[0110] The NHWOA algorithm introduces a niche strategy to divide the population into 4 niches (8 individuals in each niche). Each niche is updated independently (based on local optima). Every 10 iterations, individuals are randomly redistributed (to improve population diversity).

[0111] Four niches: The number of niches introduced by the NHWOA algorithm is updated independently through population partitioning, thereby improving the algorithm's exploration capabilities.

[0112] 8 individuals: The number of individuals in each niche population, with a total population size of [missing information]. .

[0113] 10 iterations: The interval number of iterations for random redistribution of individuals in niche habitats. This redistribution of individuals enhances population diversity and avoids premature convergence.

[0114] The NHWOA algorithm uses heuristic parameter tuning, in the early stages of iteration ( )set up (Focusing on encircling prey and random searching, enhancing exploration capabilities), later iterations ( )set up (Focusing on spiral attacks to improve accuracy). Calculated using Levy flight disturbance formulas. (s is the step size vector, , , The old individual position vector is randomly perturbed to generate a new individual position vector, thus avoiding population stagnation.

[0115] The current iteration number of the algorithm is the core basis for heuristic parameter adjustment. The maximum number of iterations for the algorithm, with a value of 500, is the benchmark for dividing the iteration phases. The threshold for dividing the iteration phase is used to distinguish between the early and late stages of algorithm iteration. : Heuristic probability parameters of the NHWOA algorithm, used to balance the algorithm's exploration and exploitation capabilities. Early stage of iteration ( The heuristic probability parameter value of ) is the first probability parameter value, which focuses on surrounding the prey and random search, and enhances the algorithm's global exploration capability. Later stages of iteration ( The heuristic probability parameter value, i.e., the second probability parameter value, focuses on spiral attacks to improve the accuracy of local exploitation in the algorithm. Levy flight perturbation: a perturbation mechanism introduced by the NHWOA algorithm to avoid population stagnation and improve the algorithm's ability to escape local optima. The new individual position vector obtained after Levy flight perturbation corresponds to a new solution for multi-source feature weights and risk thresholds. : The old individual position vector before the perturbation, corresponding to the current solution of multi-source feature weights and risk threshold. s: The step size vector of Levy flight, controlling the perturbation amplitude. The shape parameter of Levy flight, with a value of 1.5, determines the shape of the Levy distribution. Levy distribution function, calculated as follows: This provides a random step size for the perturbation. : A random variable that follows a normal distribution and satisfies . A random variable that follows a standard normal distribution and satisfies . :random variable The normal distribution variance provides parameter support for Levy flight disturbances.

[0116] The optimization objective of the NHWOA algorithm is to maximize the accuracy of pest and disease risk assessment by optimizing the weights of multi-source features. and risk threshold To improve the accuracy of risk assessment, the formula for calculating pest and disease risk assessment is as follows: ( The integrated risk of pests and diseases is quantified by weighting 15-dimensional multi-source features with optimized weights.

[0117] : Multi-source feature set Features The weighting coefficients, obtained by optimization using the NHWOA algorithm, reflect the contribution of this feature to pest and disease risk assessment. The normalization constraint on feature weights ensures that the sum of the weights of the 15 features is 1, thus guaranteeing the rationality of the risk assessment results. The pest and disease risk threshold, obtained by optimizing the NHWOA algorithm, has a value of 0.75 and is the core judgment basis for risk warning classification. The comprehensive risk value for pests and diseases is obtained by weighted summation of multiple features and their corresponding weights, and is used to assess the risk of pest and disease occurrence.

[0118] NHWOA key parameters: population size 32, maximum number of iterations 500, shrinkage factor linearly decreasing Screw coefficient .

[0119] Population size 32: The total number of individuals in the NHWOA algorithm population, consisting of 4 niches and 8 individuals in each niche.

[0120] Maximum number of iterations 500: The upper limit of the NHWOA algorithm's iterations, ensuring that the algorithm fully converges.

[0121] contraction factor The shrinkage factor of the NHWOA algorithm decreases linearly with the iteration process, and its value range is... This is used for the exploration and utilization of balancing algorithms.

[0122] Screw coefficient : The spiral coefficient of the NHWOA algorithm, which controls the shape of the spiral attack, and has a value of 1.

[0123] The computing architecture of the mushroom pest and disease early warning system adopts a combination of edge computing and cloud computing. Multi-source feature optimization and risk assessment are completed in the cloud, while preliminary image recognition is performed locally at the gateway (time-consuming). (seconds); in network latency Under the condition of seconds, the time consumption of cloud-based multi-source feature optimization and risk assessment seconds, total response time Seconds. When the optimized risk value hour, It is a disease and pest early warning trigger condition, when the optimized comprehensive risk value Greater than or equal to the risk threshold When necessary, a tiered early warning system is activated, generating tiered warnings based on disease severity:

[0124] (1) Disease level 1-2: low to medium risk level, corresponding to yellow warning, warning recommendation is to strengthen monitoring;

[0125] (2) Disease level 3: medium to high risk level, corresponding to orange alert, and the alert recommends initiating physical prevention and control measures;

[0126] (3) Disease level 4: High risk level, corresponding to a red alert. The alert recommends implementing chemical control. The alert information is pushed through a mobile terminal APP (e.g., Figure 4 (As shown).

[0127] In step S3 of this embodiment of the invention, the morphology, color, and texture features of lesions are extracted by image segmentation, and a 15-dimensional feature set is constructed by combining environmental meteorological data. The risk assessment model is optimized by using the NHWOA algorithm to realize graded early warning and precise control of pests and diseases, thereby improving the timeliness and accuracy of pest and disease identification and early warning.

[0128] S4: Based on the real-time environmental parameters of the monitoring network and the pest and disease risk level, the system generates control instructions that integrate irrigation, canopy closure adjustment and plant protection measures through decision logic, drives the corresponding equipment to execute, and then adjusts the decision according to the execution feedback to form a closed loop.

[0129] Step S4 is used to achieve comprehensive decision-making and closed-loop execution, including the following steps:

[0130] S4.1: Construction of the Decision Control Unit

[0131] The hardware uses an industrial-grade STM32F103 microcontroller (72MHz processing speed, 128KB storage capacity), integrating a data storage module (SD card, 32GB capacity) and a communication module (supporting 4G / LoRa dual-mode communication); the software is based on the FreeRTOS operating system, integrating data parsing, decision-making algorithms, command issuance and other functional modules, and all electronic devices meet the IP67 industrial protection level.

[0132] S4.2: Decision Logic Design

[0133] The decision logic design of this invention embodiment is as follows: when only a single environmental parameter deviates from a preset range, a first type of control instruction is generated and executed; when multiple environmental parameters are abnormally cross-referenced or the pest and disease risk level exceeds a preset threshold, a second type of comprehensive control instruction integrating multiple control measures is generated, and the corresponding equipment is driven to execute. Specifically:

[0134] S4.2.1: Comprehensive Regulation of Single-Parameter Anomalies

[0135] Step S4.2.1 is used to achieve rapid closed-loop control based on a single parameter threshold. It independently triggers and executes precise irrigation or forest canopy closure adjustment based on real-time soil moisture content and light intensity data. Specifically, when a single environmental parameter deviates from the suitable range (e.g., ... , When the system is in operation, control commands (start drip irrigation, set up shade nets) are generated directly according to preset rules. The command format is "actuator + operating parameters + execution duration" (e.g., "drip irrigation system + ...). +30 minutes). More specifically, it includes the following steps:

[0136] S4.2.1.1: Intelligent Irrigation Logic

[0137] Soil moisture content collected in real time by an environmental monitoring network (Percentage of field capacity), air humidity and light intensity Establish a threshold triggering mechanism. The formula for calculating the precise irrigation amount for *Gnaphalium affine* cultivation under forest cover is as follows: By taking into account the area of ​​the zone, soil characteristics, and the difference between the current moisture content and the target moisture content, the precise amount of water to be replenished can be calculated.

[0138] in, : Single precise irrigation volume in the Maocigu forest understory planting area, that is, the amount of water that the drip irrigation system needs to replenish, in L (liters). : Zoned area of ​​planting area ( (), which is the basic parameter for calculating irrigation volume, and the unit is (square meters) : Soil wetting layer depth (e.g.) The depth was set to 0.2m based on the root distribution characteristics of *Gnaphalium affine* to ensure that irrigation water reaches the root absorption area directly. Soil field water holding capacity, expressed as volumetric water content, with a range of values. It is suitable for the characteristics of sandy loam soil in karst landforms. : The percentage of soil moisture content relative to field capacity at any given time is the core basis for irrigation triggering and water volume calculation. The density of water, taking values ​​of (equivalent to) (), used to convert volume into actual water volume.

[0139] when When the water level reaches the lower limit of the suitable range, start the drip irrigation system, calculate the water replenishment requirement according to the formula, and the irrigation time for a single irrigation should not exceed 60 minutes; when When the humidity reaches the upper limit of the suitable range, open the corresponding drainage branch valve to accelerate the drainage of accumulated water; when the air humidity... And light intensity When necessary, increase the irrigation frequency appropriately (shorten the monitoring interval to 5 minutes) to prevent the leaves from losing water and wilting. : Relative humidity of air at any time, in units of It is used in conjunction with light intensity as a basis for adjusting irrigation frequency. Light intensity under the forest canopy at time t, expressed in lux.

[0140] S4.2.1.2: Dynamic Regulation of Canopy Closure

[0141] Following the principles of "low impact and reversibility priority," the light intensity under the forest canopy is monitored in real time using light sensors: when In low light conditions, selectively prune the tree layer (retaining healthy branches and removing diseased, weak, and overly dense branches), controlling the amount of pruning to a certain percentage of the total canopy size. Secondly, the interval between two pruning sessions should be no less than 30 days to avoid excessive pruning that could harm the health of the trees; when During periods of strong sunlight, prioritize setting up shade nets (shading rate...). (Reversible and easy to operate) If the deployment of the shade net is inconvenient, a small number of branches in the lower layer of trees should be retained. The adjustment cycle is to evaluate once every 7 days. After adjustment, environmental parameters are collected 3 times (10 minutes apart) to verify whether the light intensity has returned to the appropriate range. If it does not meet the standard, fine-tuning is started again (the amount of pruning is adjusted by 5%, and the shading rate of the shade net is increased or decreased by 10%).

[0142] S4.2.2: Multi-parameter anomaly integrated control:

[0143] When multiple key environmental parameter combinations deviate from their preset suitable ranges, or when the risk level of pests and diseases exceeds a preset threshold, the system will trigger a comprehensive decision-making process. These parameter combination thresholds are pre-set based on the growth physiological characteristics of *Saussurea involucrata* and the occurrence patterns of pests and diseases, aiming to identify complex environmental conditions that pose synergistic stress to plant growth. Typical abnormal conditions that trigger the second type of comprehensive control command include, but are not limited to:

[0144] (1) Water stress combination: soil moisture content Below the lower limit (e.g., <80%) and relative humidity of air Below the lower limit (e.g., <80%), this combination indicates an environment of drought, which may lead to a double obstruction of plant water metabolism.

[0145] (2) High temperature and strong light stress combination: air temperature Temperatures above the upper limit (e.g., >25℃) and forest understory light intensity Above the upper limit (e.g., >1500 lux), this combination can easily lead to leaf burn and excessive transpiration.

[0146] (3) Combinations of environments prone to disease: air temperature relative humidity of air At the same time, it remains in the high range of the appropriate range (e.g., > 22℃ and >85%), these warm and humid conditions easily induce fungal diseases.

[0147] (4) Combination of pest and disease risk with environmental factors: The comprehensive pest and disease risk value R reaches the high-risk threshold (e.g., ≥0.75) and is accompanied by an abnormality in any key environmental parameter (e.g., R ≥ 0.75 and...). >25℃), indicating that the risk of pest and disease outbreaks increases sharply under adverse environmental conditions.

[0148] Specifically, when the system detects things such as <80% and <80% (moisture stress), or R≥0.75 and When conditions such as >25℃ (a combination of pest and disease risk and high environmental temperature) meet the above logic, it is determined to be a complex scenario requiring coordinated response. At this point, the system will invoke the NHWOA-optimized weighted decision model. ( For the baseline control strategy targeting the j-th abnormal parameter, The weighting coefficients for NHFOA are dynamically optimized in real time, with the optimization objective being to minimize control costs and maximize effectiveness, in order to generate a comprehensive control scheme that integrates irrigation, canopy closure adjustment, and plant protection measures. This model achieves precise and coordinated control by balancing multiple objective requirements (e.g., while replenishing water and cooling, it also takes into account shading to reduce light stress and the timing of pesticide application).

[0149] The optimization process of the weighted decision model is as follows: each possible integrated control scheme is encoded as an individual position vector of the NHWOA algorithm, with each vector element corresponding to a weight coefficient. The optimization objective, or fitness function, is set as a function that comprehensively considers the economic cost of regulatory operations and the effect score of the regression of expected environmental parameters within the suitable range, for example... ,in , The fitness function is the balance coefficient. The NHWOA algorithm searches to make the fitness function... The weight combination with the smallest value This allows for a comprehensive decision that achieves the optimal balance between cost and effectiveness.

[0150] The soil moisture content single parameter abnormal threshold (lower limit of the suitable range) is the core condition for triggering the start of the drip irrigation system. Lux: The single-parameter abnormal threshold of light intensity (upper limit of the suitable range), the core condition for triggering the control command to set up a shade net. Actuator + Operating Parameters + Execution Time: The standardized format of the control command when a single parameter is abnormal, such as "drip irrigation system..." "Minutes" ensures that the instructions can be accurately recognized by the execution device. The operating parameters (irrigation volume per unit area) of the drip irrigation system are the preset water replenishment standards when a single parameter is abnormal. and An abnormal situation where soil moisture content and relative humidity intersect triggers multi-parameter integrated regulation. and The second scenario is when the combined risk value of pests and diseases and the abnormal conditions of air temperature cross, triggering multi-parameter integrated regulation. The comprehensive decision value under multi-parameter anomalies is obtained by weighted summation of the decision results of single parameters and is used to generate a comprehensive control plan. The number of abnormal parameters involved in the comprehensive decision-making process, i.e., the total number of single-parameter decision results that need to be weighted. : No. A single parameter decision result The weighting coefficients are dynamically optimized in real time by the NHWOA algorithm, with the optimization goal of minimizing control costs and achieving optimal results. : No. The decision result when a single parameter deviates from the appropriate range represents the basic control strategy (such as water replenishment, shading, pesticide application, etc.) corresponding to the abnormality of the parameter.

[0151] S4.3: Closed-loop control execution

[0152] The generated decision commands are sent to the corresponding drip irrigation systems, drainage systems, and stand control devices for execution. Stand control prioritizes manual pruning guidance (low cost, easy operation), with robotic pruning arms used as an experimental solution for small-scale demonstration. After execution, the drip irrigation systems, drainage systems, and stand control devices return status signals (e.g., "drip irrigation complete," "valve closed") to the decision unit. The decision unit updates the database and system status based on the returned signals, continuously monitors parameter changes, and determines whether the updated environmental parameters have returned to the suitable range. If the parameters have not returned to the suitable range, the decision process is re-triggered to generate new control commands, achieving fully automated closed-loop management of monitoring-decision-execution-feedback. For example, in single-parameter abnormal control, after each irrigation or canopy closure adjustment, environmental parameters are collected three times consecutively (10-minute intervals) to verify whether the parameters have returned to the suitable range. If the standard is not met, fine-tuning (increasing or decreasing irrigation volume) is restarted. Adjustment of pruning amount This forms a closed-loop logic of "monitoring-regulation-verification" to ensure that water and light environments are adapted to the growth of Sagittaria trifolia.

[0153] In summary, the intelligent ecological cultivation method for *Gnaphalium affine* under forest cover based on environmental monitoring provided in the above embodiments of the present invention extracts the morphological, color, and texture features of lesions through image segmentation, constructs a 15-dimensional feature set by combining environmental meteorological data, and optimizes the risk assessment model using the NHFOA algorithm to achieve graded early warning and precise control of pests and diseases, thereby improving the timeliness and accuracy of pest and disease identification and early warning. A complete technical system of "multi-parameter monitoring - multi-source feature fusion - intelligent early warning - collaborative regulation" is constructed to achieve real-time perception and precise assessment of the microenvironment (light, temperature, humidity, and soil parameters) and pest and disease risks under the *Gnaphalium affine* forest cover. A collaborative regulation mechanism for irrigation and canopy density is established, and a decision-making model optimized by NHFOA is used to achieve dynamic adaptation of water supply and light conditions, solving the cultivation pain points of "uneven drought and flood" and "unbalanced light" in *Gnaphalium affine* under karst topographical habitats.

[0154] like Figure 5 As shown, the algorithm of this invention (NHWOA) outperforms traditional algorithms in terms of recognition accuracy, reaching over 90%.

[0155] It should also be noted that, in the embodiments of this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0156] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in the embodiments of this application may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown in this application, but is to be accorded the widest scope consistent with the principles and novel features disclosed in the embodiments of this application.

Claims

1. A method for intelligent ecological cultivation of *Gnaphalium affine* under forest cover based on environmental monitoring, characterized in that: include: S1: Determine the appropriate target canopy closure range based on the forest stand type of the target planting area, and set up a monitoring network to continuously acquire environmental parameters and leaf images of the hairy taro growth. S2: Implement the deployment of forest understory soil ecological improvement and water regulation infrastructure, including: standardizing the land preparation of the planting area, applying fully decomposed organic fertilizer and introducing earthworms for ecological improvement; constructing a layered drainage network consisting of main ditches, branch ditches and furrows, and simultaneously deploying an intelligent drip irrigation network with pressure sensing and filtration functions. S3: Based on the leaf image, identify pests and diseases and locate lesions; segment the lesion area, extract its morphological, color and texture features and fuse them with the corresponding spatiotemporal environmental parameters to construct a multi-source feature set; use the improved microhabitat hybrid heuristic whale optimization algorithm to optimize the risk assessment model constructed based on the multi-source feature set, and output the pest and disease risk level based on the optimized risk assessment model; The improved niche hybrid heuristic whale optimization algorithm optimizes the risk assessment model, including: A1. Initialization: Set algorithm parameters, including population size and maximum number of iterations, and divide the population into multiple independent microhabitats, each microhabitat including several population individuals; A2. Iterative optimization: Each niche is independently updated based on its local optimum. Update strategies include prey encirclement, random search, and spiral attack. A3. Diversity maintenance: When the preset redistribution period is reached, individuals in all niches are randomly redistributed; A4. Perturbation and Bounceout: During the iteration process, random perturbations based on Lévy flight are applied to the individual positions; A5. Parameter Adaptation: In the early stage of iteration, the first probability parameter value is used to focus on global exploration, and in the later stage of iteration, the second probability parameter value is used to focus on local development; Repeat steps A2 to A5 until the maximum number of iterations or the convergence condition is met, and output the optimal feature weight set and risk threshold. S4: Make decisions based on the real-time environmental parameters of the monitoring network and the pest and disease risk level, generate control instructions that integrate irrigation, canopy closure adjustment and plant protection measures based on the decision logic, and drive the corresponding equipment to execute them. Then adjust the decision according to the execution feedback to form a closed loop.

2. The method according to claim 1, characterized in that, In step S1, the forest stand types include coniferous forest, broad-leaved forest and mixed forest, and the corresponding target canopy closure ranges are 0.6-0.7, 0.5-0.6 and 0.55-0.65, respectively.

3. The method according to claim 1, characterized in that, In step S3, a multi-source feature set is constructed, including: The YOLOv8 target detection model was used to perform preliminary identification on the leaf images, and the category, confidence level and bounding box coordinates of the lesions were output. Based on the bounding box coordinates, the lesion region is located, and the Mask R-CNN semantic segmentation model is used to perform pixel-level segmentation of the lesion region to obtain an accurate binary lesion mask. Based on the binarized lesion mask, the morphological features, color features, and texture features of the lesions are calculated and extracted respectively; wherein, the morphological features include at least the lesion area, perimeter, and roundness; the color features include at least the mean values ​​of each channel of the lesion region in the RGB color space and the peak values ​​of the histogram in the HSV color space; the texture features include at least the energy and entropy calculated based on the gray-level co-occurrence matrix; The extracted morphological, color, and texture features are fused with four environmental parameters acquired simultaneously during the acquisition of the leaf images: air temperature, relative humidity, soil moisture content, and forest light intensity, to form a 15-dimensional multi-source feature vector for risk assessment.

4. The method according to claim 1, characterized in that, The goal of optimizing the risk assessment model is to maximize the accuracy of pest and disease risk assessment. This is achieved through synchronous iterative optimization of the weights of each feature in the risk assessment model. and risk assessment threshold To achieve this, the formula for calculating the integrated pest and disease risk value R is: ； in, For the first in the multi-source feature set One characteristic, The corresponding feature weights obtained by optimizing the NHWOA algorithm, and where n is the total number of features; When the comprehensive risk value R of pests and diseases is greater than or equal to the preset risk threshold At that time, an early warning is triggered.

5. The method according to claim 1, characterized in that, In step S4, the decision logic is configured as follows: When a single environmental parameter deviates from the preset range, a first-type control command is generated and executed. When multiple environmental parameters cross abnormally or the pest and disease risk level exceeds a preset threshold, a second type of comprehensive control command integrating multiple control measures is generated and the corresponding equipment is driven to execute.

6. The method according to claim 5, characterized in that, When only a single environmental parameter deviates from the preset range, the generation and execution of a first type of control instruction includes: Based on a single-parameter threshold-based rapid closed-loop control, precision irrigation or forest canopy closure adjustment is independently triggered and executed according to real-time soil moisture content and light intensity data. The condition for triggering precision irrigation is that the percentage of soil moisture content to field capacity is less than 80%. The precision irrigation amount Q is calculated based on the following formula: ； Where Q is the irrigation amount, Where is the area of ​​the zone, and h is the depth of the soil wetting layer. It refers to the soil field water holding capacity. This represents the current soil moisture content as a percentage of field capacity. This is the density of water.

7. The method according to claim 6, characterized in that, The adjustment of forest canopy density includes: when the light intensity under the forest canopy is adjusted. At that time, selective pruning is carried out on the tree layer; when the light intensity under the forest canopy is high... When doing so, prioritize setting up shade nets or preserving the lower branches of trees.

8. The method according to claim 5, characterized in that, When generating the second type of integrated control command, the following decision model is invoked to perform multi-objective trade-offs: ;in, This is the comprehensive decision value when multiple parameters are abnormal; For the baseline control strategy targeting the j-th abnormal parameter, These are the corresponding weighting coefficients.

9. The method according to claim 1, characterized in that, Step S4 includes: The generated control commands are sent to the corresponding drip irrigation system, drainage system or forest stand control device for execution. After execution, a status confirmation signal is returned to the decision-making unit. The decision-making unit updates the system status based on the returned signal and determines whether the updated environmental parameters have returned to the appropriate range. If they have not returned, the decision-making process is retried to generate new control commands, thereby realizing closed-loop control of monitoring-decision-execution-feedback.

Images