Cotton field growth state evaluation and prediction system based on unmanned aerial vehicle rgb image

Abstract

The application provides a cotton field growth state evaluation and prediction system based on a UAV RGB image, comprising the following modules connected with each other: a data acquisition module for acquiring initial image data; a data processing module for denoising, distortion correction, ortho-mosaic and light correction to obtain standard image data; an image mining module for image segmentation and target detection, and further extraction of cotton contour features and growth indexes; a comprehensive evaluation module for constructing a growth state evaluation model to obtain a comprehensive growth condition score and evaluation indexes; a growth prediction module for introducing yield-related factors to perform yield prediction and growth anomaly early warning; and a real-time monitoring and decision support module for visual display of the evaluation and prediction results and automatic generation of decision support. The application can comprehensively analyze the cotton field growth, improve the comprehensiveness, accuracy and efficiency of evaluation and prediction, and further perform real-time monitoring and intelligent decision on the whole cotton growth process, thereby improving the efficiency and accuracy of agricultural management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of plant growth evaluation, in particular to a cotton field growth state evaluation and prediction system based on an unmanned aerial vehicle RGB image. BACKGROUND

[0002] Cotton plants are important economic crops worldwide and are the main raw material for the textile industry, widely used in the production of clothing, home textiles and other products, supporting the global textile industry chain. By studying the comprehensive growth status of cotton, planting management measures can be optimized, cotton yield and fiber quality can be improved, and spatial and temporal variation characteristics of cotton growth can be understood, which helps to achieve precise fertilization, irrigation and pest control, and reduces agricultural input costs.

[0003] With the development of science and technology, existing methods and technologies for evaluating the comprehensive growth of cotton plants include but are not limited to hyperspectral imaging, Internet of Things sensors, etc., but most of these technologies have high costs and limited data processing, making them difficult to be widely applied. Using unmanned aerial vehicle RGB images to evaluate cotton growth can reduce costs, and unmanned aerial vehicles have high flexibility and can quickly cover large areas of planting areas. However, existing technologies for evaluating cotton growth using unmanned aerial vehicle RGB images still have the following problems: 1) high data processing complexity, large amount of RGB image data, low processing efficiency, and image processing algorithms sensitive to light conditions, which may result in unstable results; 2) lack of comprehensive analysis of multiple factors: existing researches mostly focus on a single indicator (such as canopy coverage), and lack comprehensive analysis of multiple factors (such as soil and weather); 3) insufficient real-time performance: there is a time delay between data collection and analysis, making it difficult to achieve real-time monitoring and dynamic evaluation. SUMMARY

[0004] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a cotton field growth state evaluation and prediction system based on an unmanned aerial vehicle RGB image, which can combine multiple factors for comprehensive analysis of cotton growth, improve the comprehensiveness, accuracy and efficiency of evaluation and prediction, and also enable real-time monitoring and intelligent decision-making of the entire cotton growth process, thereby improving the efficiency and accuracy of agricultural management and promoting cotton precision agriculture to a new stage of digitization and intelligentization.

[0005] To achieve the above-mentioned purpose, the present application provides the following scheme: a cotton field growth state evaluation and prediction system based on an unmanned aerial vehicle RGB image, comprising:

[0006] a data acquisition module for selecting an unmanned aerial vehicle, mounting an RGB camera on the selected unmanned aerial vehicle, pre-setting an unmanned aerial vehicle route and image acquisition parameters according to the terrain of a land plot, and then performing data acquisition to obtain initial image data;

[0007] a data processing module, configured to denoise, correct distortion, ortho-mosaic and correct illumination of the initial image data to obtain standard image data;

[0008] an image mining module, configured to perform image segmentation and target detection on the standard image data to obtain cotton target information, and extract canopy coverage, plant height, color features and vegetation index, and texture and spatial structure features of the cotton target information to obtain cotton profile features and growth indexes;

[0009] a comprehensive evaluation module, configured to introduce environmental data, train a machine learning model by using the environmental data and the growth indexes to obtain a growth state evaluation model, and obtain a comprehensive growth condition score and an evaluation index by using the growth state evaluation model;

[0010] a growth prediction module, configured to introduce yield-related factors based on the growth state evaluation model to perform yield prediction, and preset a growth anomaly threshold to perform growth anomaly early warning;

[0011] a real-time monitoring and decision support module, configured to use a cloud service platform to perform distributed storage and management of data, use WebGIS technology to visually display comprehensive evaluation results and growth prediction results of each cotton field, and automatically generate decision support according to the comprehensive evaluation results and the growth prediction results, and perform automatic management according to the decision support;

[0012] The data collection module, the data processing module, the image mining module, the comprehensive evaluation module, the growth prediction module and the real-time monitoring and decision support module are connected to each other.

[0013] Optionally, the data collection module comprises:

[0014] a hardware configuration unit, configured to select a UAV, and mount an RGB camera and a three-axis gimbal matched with the RGB camera on the selected UAV to realize anti-shake performance of the camera;

[0015] a flight path planning unit, configured to automatically generate a flight path by using a UAV flight path planning software, and set a front overlap degree and a side overlap degree of an image;

[0016] a parameter setting unit, configured to set a flight height and a resolution of the UAV according to a cotton growth stage, and then perform data collection to obtain initial image data.

[0017] Optionally, the data processing module comprises:

[0018] a denoising and distortion correction unit, configured to perform denoising processing on the initial image data by using an automatic batch processing tool and a denoising algorithm, and perform distortion correction on an image edge by using camera calibration parameters;

[0019] Ortho-mosaic unit, configured to perform feature point matching and image stitching on the overlapping images in the initial image data by using aerial survey processing software, and output an ortho-image with geographic coordinate system information;

[0020] Illumination correction unit, configured to deploy a gray scale board inside a flight area of the unmanned aerial vehicle, record a standard reflection value, perform color correction on the initial image data based on the standard reflection value, and integrate a radiation correction algorithm in the aerial survey processing software to standardize image brightness and image color.

[0021] Optionally, the image mining module comprises:

[0022] Data labeling unit, configured to perform pixel-level labeling by using a labeling tool based on the standard image data, generate a segmentation label of cotton plants and soil background, and perform bounding box labeling on single cotton plants and pest and disease areas;

[0023] Image segmentation unit, configured to select a convolutional neural network, introduce an attention mechanism, a multi-scale feature extraction module and a depth separable convolution into the selected convolutional neural network to obtain a first initial model, train the first initial model by using a deep learning framework to obtain a segmentation model, and segment the standard image data by using the segmentation model to output a binary image; wherein the binary image comprises overall distribution information of a cotton canopy.

[0024] Target detection unit, configured to obtain a target detection model by combining YOLOv5 and multi-task learning, detect the binary image by using the target detection model, and output position information; wherein the position information comprises positions of single cotton plants or pest and disease areas.

[0025] Result fusion unit, configured to obtain cotton target information by combining the binary image and the position information.

[0026] Outline feature and growth index extraction unit, configured to extract canopy coverage, plant height, color features and vegetation indexes, and texture and spatial structure features of the cotton target information to obtain outline features and growth indexes of the cotton target information.

[0027] Optionally, the outline feature and growth index extraction unit comprises:

[0028] Canopy coverage calculation sub-unit, configured to count a proportion of cotton canopy pixels in the binary image to obtain canopy coverage, and evaluate whether a growth process is normal according to the canopy coverage in combination with a cotton growth period.

[0029] a plant height estimation subunit configured to estimate plant height in combination with the orthographic image and the digital surface model, and output a spatial distribution map of the plant height to identify abnormal growth areas;

[0030] a texture and spatial structure feature extraction subunit configured to extract texture features based on the standard image data using a gray level co-occurrence matrix, and calculate population density and plant spacing of the cotton plants to obtain spatial structure features;

[0031] a color feature and vegetation index calculation subunit configured to extract color features of RGB bands in the standard image data, design an RGB vegetation index in combination with the texture features and the color features, and evaluate the health status of the cotton and identify disease and pest areas according to a spatial distribution of the RGB vegetation index.

[0032] Optionally, the comprehensive evaluation module comprises:

[0033] a data fusion unit configured to obtain soil data, meteorological data and geographic information data of a cotton field to obtain environmental data, perform spatial alignment and time alignment on the environmental data and the standard image data, and fuse the environmental data and the extracted growth indicators to obtain a comprehensive data set;

[0034] a model construction unit configured to construct an initial model using an ensemble learning method, set input variables and output variables of the initial model, and perform model training using the comprehensive data set to obtain a growth state evaluation model; the input variables comprise canopy coverage, plant height, vegetation index, texture features and environmental data, and the output variables comprise a comprehensive growth condition score, a health index and a stress index;

[0035] an indicator quantification unit configured to generate a spatial distribution map according to the comprehensive growth condition score, and obtain evaluation indicators in combination with the health index and the stress index to perform comprehensive evaluation.

[0036] Optionally, an expression of the comprehensive growth condition score is:

[0037] S = ω1·C + ω2·H + ω3·VI + ω4·T

[0038] wherein S is the comprehensive growth condition score, C is the canopy coverage, H is the plant height, VI is the vegetation index, T is the texture features, and ω1, ω2, ω3 and ω4 are weights of the indicators;

[0039] an expression of the health index is:

[0040] HI = α·VI + β·T

[0041] wherein HI is the health index, and α and β are weight parameters.

[0042] The expression of the adversity index is:

[0043] AI = γ1 · (1 - E soil ) + γ2 · |E temp -T opt |

[0044] Wherein, AI is the adversity index, E soil is the soil humidity, E temp is the temperature, T opt is the optimal growth temperature of cotton, γ1 and γ2 are weight parameters.

[0045] Optionally, the growth prediction module comprises:

[0046] A yield prediction unit is configured to fuse the canopy coverage, the plant height, the vegetation index, the texture feature, the environmental data and a yield-related factor to obtain a yield prediction dataset, to train a yield prediction model by using the yield prediction dataset, and to output a single-plant yield prediction value and a unit-area yield prediction value by using the yield prediction model; wherein the yield-related factor comprises a plant density and a boll weight, and the yield prediction model introduces an environmental correction factor for adjusting the output value;

[0047] A growth anomaly unit is configured to set a comprehensive growth condition score threshold, a health index threshold and an adversity index threshold according to the output value of the growth state evaluation model, and to determine whether a cotton plot is an abnormal area according to the set thresholds, and if so, to automatically generate a warning information.

[0048] Optionally, the expression of the single-plant yield prediction value is:

[0049] Y single = B · W boll

[0050] Wherein, Y single is the single-plant yield prediction value, B is the boll weight, and W boll is the single-boll weight.

[0051] The expression of the unit-area yield prediction value is:

[0052] Y yield = D · Y single · (1 - k1 · |E soil -E opt | - k2 · |E temp -T opt |)

[0053] Wherein, Y yield is the unit-area yield prediction value, D is the plant density, Eopt T is the optimal soil moisture, opt T is the optimal growth temperature, k1, k2 are environmental correction coefficients.

[0054] Optionally, the real-time monitoring and decision support module comprises:

[0055] The data storage unit is configured to store structured data based on a cloud service platform using a relational database, manage unstructured data using object storage, and perform real-time data access through an API interface.

[0056] The monitoring and visualization unit is configured to develop a responsive interface supporting multi-terminal access using a front-end framework, and to display the processing results of the image mining module, the comprehensive evaluation module and the growth prediction module in real time on the interface in the form of map overlay, plot view and heat map using WebGIS technology.

[0057] The decision support unit is configured to generate fertilization and irrigation recommendations, pest control and supervision recommendations, climate change trend analysis and pest regularity analysis based on the processing results of the image mining module, the comprehensive evaluation module and the growth prediction module.

[0058] The present application provides a cotton field growth state evaluation and prediction system based on unmanned aerial vehicle RGB images, and discloses the following technical effects:

[0059] 1. Efficient data processing: The present application can truly reflect near-infrared or hyperspectral results by automatic data processing and sufficient image mining, and can comprehensively and accurately reflect cotton growth conditions by multi-features such as canopy coverage, plant height, color features, vegetation index, and texture and spatial structure features, providing a reliable data basis for subsequent cotton growth evaluation and prediction.

[0060] 2. High comprehensive evaluation accuracy: The present application can obtain multi-dimensional comprehensive evaluation data by fusing unmanned aerial vehicle image data with external environmental data such as soil and weather in space and time dimensions, to construct a growth state evaluation model, obtain a quantitative comprehensive growth condition score and a health index, not only comprehensively revealing the comprehensive growth condition of cotton, but also improving the scientificity and reliability of the evaluation.

[0061] 3. Real-time monitoring and intelligent decision making: The present application can realize efficient data storage and management by building a cloud service platform, ensure real-time data uploading, display comprehensive evaluation results and growth prediction results in real time through a visual interface, realize real-time monitoring and multi-terminal access of data, and automatically generate fertilization and irrigation recommendations, pest control and supervision recommendations, climate change trend analysis and pest regularity analysis, etc., to improve the intelligent level of agricultural management.

[0062] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a schematic diagram of the system architecture provided in an embodiment of the present invention;

[0065] Figure 2 This is a schematic diagram of the image mining process provided in an embodiment of the present invention;

[0066] Figure 3 This is a schematic diagram of the evaluation and prediction process provided in an embodiment of the present invention. Detailed Implementation

[0067] 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.

[0068] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0069] like Figure 1 As shown, the present invention provides a cotton field growth status assessment and prediction system based on UAV RGB images, including interconnected data acquisition module, data processing module, image mining module, comprehensive assessment module, growth prediction module and real-time monitoring and decision support module.

[0070] 1. Data Acquisition Module

[0071] The system is used to select a drone, which is equipped with an RGB camera. Based on the terrain, the drone's flight path and image acquisition parameters are preset before data acquisition to obtain initial image data. The data acquisition module includes:

[0072] 1.1 Hardware Configuration Unit

[0073] For selecting a drone, and mounting an RGB camera and a three-axis gimbal compatible with the RGB camera on the selected drone to achieve anti-shake performance of the camera and reduce image blur caused by vibration.

[0074] 1.2 Route planning unit

[0075] For task design using drone route planning software (such as DJI Pilot, Pix4D capture, UGCS). Input cotton field boundary and terrain data to automatically generate flight paths covering the entire area.

[0076] Set the forward overlap and lateral overlap of the image. The forward overlap is set to 70%-80% to ensure sufficient overlap area between images for stitching. The lateral overlap is set to 60%-70% to avoid blank areas when stitching images.

[0077] 1.3 Parameter setting unit

[0078] For setting the flight height and resolution of the drone according to the growth stage of cotton, and then collecting data to obtain initial image data.

[0079] 2. Data processing module

[0080] For denoising, distortion correction, ortho-mosaic and illumination correction of the initial image data to obtain standard image data; the data processing module includes:

[0081] 2.1 Denoising and distortion correction unit

[0082] For denoising the initial image data using automatic batch processing tools (such as OpenCV or ImageJ) and denoising algorithms (such as mean filtering, Gaussian filtering) to reduce random noise interference on image features.

[0083] Use camera calibration parameters (such as focal length, distortion coefficient) to correct the distortion of image edges and correct the image edge curvature caused by lens optical properties. Open source tools (such as the `cv2.undistort()` function of OpenCV) can be used to achieve automatic correction.

[0084] 2.2 Ortho-mosaic unit

[0085] For feature point matching and image stitching of overlapping images in the initial image data using photogrammetry processing software (Pix4D, Agisoft Metashape, etc.), outputting ortho-images with geographic coordinate system information.

[0086] 2.3 Illumination correction unit

[0087] For deploying a gray scale board inside the flight area of the unmanned aerial vehicle, recording a standard reflectance value, color correcting the initial image data based on the standard reflectance value, reducing light difference, and integrating a radiation correction algorithm (such as Empirical Line Calibration, ELC) in the aerial survey processing software to standardize image brightness and image color.

[0088] 3. image mining module

[0089] As shown in Figure 2 , for image segmentation and target detection of the standard image data, to obtain cotton target information, and then extract the cotton target information of canopy coverage, plant height, color features, vegetation index, and texture and spatial structure features, to obtain cotton profile features and growth indicators; the image mining module comprises:

[0090] 3.1 data labeling unit

[0091] For pixel-level labeling based on the standard image data using labeling tools (such as Labelme, LabelImg), to generate segmentation labels of cotton plants and soil background, and to label the boundary boxes of individual cotton plants and pest and disease areas.

[0092] 3.2 image segmentation unit

[0093] For selecting a convolutional neural network (such as MobileNet U-Net), introducing attention mechanisms (such as SE modules or CBAM modules), multi-scale feature extraction modules, and depth separable convolutions (to reduce parameter quantity and computational load) in the selected convolutional neural network, to obtain a first initial model, and then training the first initial model using a deep learning framework to obtain a segmentation model, using the segmentation model to segment the standard image data, and outputting a binary image (cotton plant area is 1 and soil background is 0), to provide a basis for subsequent feature extraction; wherein the binary image includes overall distribution information of the cotton canopy.

[0094] 3.3 target detection unit

[0095] For combining YOLOv5 and multi-task learning (outputting cotton plant positions and pest and disease areas simultaneously in target detection to improve detection efficiency), to obtain a target detection model, using the target detection model to detect the binary image, and outputting position information; wherein the position information includes the positions of individual cotton plants or pest and disease areas.

[0096] 3.4 result fusion unit

[0097] For combining the binary image and the position information, to obtain cotton target information.

[0098] 3.5 Profile feature and growth indicator extraction unit

[0099] The profile feature and growth indicator extraction unit is configured to extract canopy coverage, plant height, color feature and vegetation index, and texture and spatial structure feature of the cotton target information, and obtain the profile feature and growth indicator of the cotton target information.

[0100] The profile feature and growth indicator extraction unit comprises:

[0101] 3.51 Canopy coverage calculation subunit

[0102] The canopy coverage calculation subunit is configured to count the proportion of cotton canopy pixels in the binary image, and obtain the canopy coverage. According to the canopy coverage, the growth process is evaluated in combination with the cotton growth period (seedling stage, budding stage, flowering and bolling stage, and boll opening stage) to determine whether the growth process is normal.

[0103] 3.52 Plant height estimation subunit

[0104] The plant height estimation subunit is configured to estimate the plant height in combination with the orthophoto image and the digital surface model, and output a spatial distribution map of the plant height to identify abnormal growth areas (such as excessively high or low plants).

[0105] 3.53 Texture and spatial structure feature extraction subunit

[0106] The texture and spatial structure feature extraction subunit is configured to extract texture features based on the standard image data using a gray level co-occurrence matrix, and calculate the population density (the number of cotton plants per unit area) and the inter-plant distance (the uniformity of the distribution of cotton plants) of the cotton plants to obtain the spatial structure feature.

[0107] 3.54 Color feature and vegetation index calculation subunit

[0108] The color feature and vegetation index calculation subunit is configured to extract color features (such as average brightness and green component) of the RGB waveband in the standard image data, combine the texture features and the color features, design an RGB vegetation index, and evaluate the health status of the cotton and identify possible disease and pest areas according to the spatial distribution of the RGB vegetation index.

[0109] 4. Comprehensive evaluation module

[0110] As shown in Figure 3 , the comprehensive evaluation module is configured to introduce environmental data, train a machine learning model using the environmental data and the growth indicators, obtain a growth state evaluation model, and obtain a comprehensive growth condition score and evaluation indicators using the growth state evaluation model.

[0111] 4.1 Data fusion unit

[0112] The soil data, meteorological data and geographic information data of the cotton field are acquired to obtain environmental data, the environmental data is spatially and temporally aligned with the standard image data, and the environmental data is fused with the extracted growth indicators to obtain a comprehensive data set.

[0113] Spatial alignment: Align the environmental data with the geographic coordinate system of the UAV image (such as WGS84). Use GIS platforms (such as ArcGIS, QGIS) for spatial overlay to map soil and meteorological data to cotton field areas to generate multi-layer spatial information maps.

[0114] Time alignment: Unify the time stamp and synchronize the environmental data with the UAV flight time to ensure the timeliness of the data.

[0115] When data fusion, in the spatial dimension, is fused by plot or pixel level. In the time dimension, it is aggregated by day, week, month, etc.

[0116] 4.2 Model construction unit

[0117] For using integrated learning methods (such as Random Forest, XGBoost) to construct an initial model, setting the input variables and output variables of the initial model, and using the comprehensive data set for model training to obtain a growth state evaluation model; the input variables include canopy coverage, plant height, vegetation index, texture features and environmental data (soil moisture, nutrients, meteorological data), and the output variables include comprehensive growth condition score (0-100 points), health index (reflecting the degree of plant health) and adversity index (reflecting the degree of drought, disease and pest adversity influence).

[0118] The expression of the comprehensive growth condition score is:

[0119] S = ω1·C + ω2·H + ω3·VI + ω4·T

[0120] Where S is the comprehensive growth condition score, C is the canopy coverage, H is the plant height, VI is the vegetation index, T is the texture feature, and ω1, ω2, ω3, ω4 are the weights of each indicator.

[0121] The expression of the health index is:

[0122] HI = α·VI + β·T

[0123] Where HI is the health index, and α and β are weight parameters.

[0124] The expression of the adversity index is:

[0125] AI = γ1·(1-E soil )+γ2·|E temp -Topt |

[0126] wherein, AI is adversity index, E soil is soil humidity, E temp is temperature, T opt is optimal growth temperature of cotton, γ1, γ2 are weight parameters.

[0127] 4.33 index quantification unit

[0128] for generating a spatial distribution map according to the comprehensive growth condition score, and combining the health index and adversity index to obtain an evaluation index for comprehensive evaluation.

[0129] 5. growth prediction module

[0130] As Figure 3 shown, for introducing yield-related factors based on the growth state evaluation model, performing yield prediction, and presetting a growth anomaly threshold to perform growth anomaly early warning; the growth prediction module comprises:

[0131] 5.1 yield prediction unit

[0132] As Figure 3 shown, for fusing the canopy coverage, the plant height, the vegetation index, the texture feature, the environmental data and the yield-related factors to obtain a yield prediction dataset, performing model training using the yield prediction dataset to obtain a yield prediction model, and outputting a single-plant yield prediction value and a unit-area yield prediction value using the yield prediction model; wherein the yield-related factors include plant density and boll weight, and the yield prediction model introduces an environmental correction factor for adjusting the output value.

[0133] The expression of the single-plant yield prediction value is:

[0134] Y single =B·W boll

[0135] wherein, Y single is a single-plant yield prediction value, B is boll weight, and W boll is single-boll weight.

[0136] The expression of the unit-area yield prediction value is:

[0137] Y yield =D·Y single ·(1-k1·|E soil -E opt |-k2·|E temp -T opt |)

[0138] wherein, Yyield Y is the yield prediction value per unit area, D is the plant density, E opt Yopt is the optimal soil moisture, T opt Yopt is the optimal growth temperature, k1 and k2 are environmental correction coefficients.

[0139] 5.2 Abnormal growth unit

[0140] For the output value according to the growth state evaluation model, set the comprehensive growth condition score threshold, health index threshold and adversity index threshold, and then judge whether the cotton plot is an abnormal area according to the set threshold. If so, automatically generate a warning message. The warning message includes the location (latitude and longitude) of the abnormal area, the analysis of the abnormal reason (such as drought, disease and insect pests), the prevention and control suggestions (such as irrigation, fertilization and pesticide spraying).

[0141] 6. Real-time monitoring and decision support module

[0142] For distributed storage and management of data using cloud service platform, visualization of comprehensive evaluation results and growth prediction results of each cotton plot using WebGIS technology, and automatic generation of decision support according to comprehensive evaluation results and growth prediction results, and automatic management according to decision support; the real-time monitoring and decision support module comprises:

[0143] 6.1 Data storage unit

[0144] For storing structured data (such as health index, yield prediction result) based on cloud service platform (such as AWS, Google Cloud, Aliyun), using relational database, managing unstructured data (such as unmanned aerial vehicle image, ortho mosaic) using object storage, and accessing real-time data through API interface.

[0145] 6.2 Monitoring and visualization unit

[0146] For developing a responsive interface supporting multi-terminal access using a front-end framework, and using WebGIS technology to display the processing results of the image mining module, the comprehensive evaluation module and the growth prediction module on the interface in the form of map overlay, plot view and heat map.

[0147] Map overlay: use open source GIS framework (such as Leaflet, OpenLayers) or commercial platform (such as ArcGIS Online) to realize the function of map overlay. Show multiple layers of information (such as health index, canopy coverage, soil moisture, disease and insect pest distribution) on the map.

[0148] Plot view: show growth data by plot, and click the plot to view detailed information (such as historical growth trend, predicted yield).

[0149] Heat map: generate spatial heat map of health index, adversity index, intuitively show abnormal growth area.

[0150] 6.3 Decision support unit

[0151] For generating fertilization and irrigation suggestions, disease and pest control supervision suggestions, climate change trend analysis (analyzing the long-term impact of climate change on cotton growth, such as the impact of increased drought frequency on yield), and disease and pest regularity analysis (using data mining algorithms such as association rule mining to discover the time and environmental conditions of disease and pest occurrence) based on the processing results of the image mining module, the comprehensive evaluation module and the growth prediction module.

[0152] Therefore, the present application provides a cotton field growth state evaluation and prediction system based on unmanned aerial vehicle RGB images, which can combine various factors to comprehensively analyze cotton growth, improve the comprehensiveness, accuracy and efficiency of evaluation and prediction, and also can monitor and make intelligent decisions in real time during the whole process of cotton growth, improve the efficiency and accuracy of agricultural management, and promote cotton precision agriculture to a new stage of digitization and intelligentization.

[0153] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0154] The principles and implementation modes of the present application are described by applying specific examples in this specification. The above description of the embodiments is only to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A cotton field growth status assessment and prediction system based on UAV RGB images, characterized in that, include: The data acquisition module is used to select a drone, mount an RGB camera on the selected drone, preset the drone flight path and image acquisition parameters according to the terrain, and then collect data to obtain initial image data. The data processing module is used to perform denoising, distortion correction, orthophoto stitching, and illumination correction on the initial image data to obtain standard image data; The image mining module is used to perform image segmentation and target detection on the standard image data to obtain cotton target information. Then, the canopy coverage, plant height, color features and vegetation index, as well as texture and spatial structure features of the cotton target information are extracted to obtain cotton profile features and growth indicators. A comprehensive evaluation module is used to input environmental data, train a machine learning model using the environmental data and the growth indicators to obtain a growth status evaluation model, and then use the growth status evaluation model to obtain a comprehensive growth status score and evaluation index; the comprehensive evaluation module includes: The data fusion unit is used to acquire soil data, meteorological data, and geographic information data of cotton fields to obtain environmental data. The environmental data is spatially and temporally aligned with the standard image data, and then the environmental data is fused with the extracted growth indicators to obtain a comprehensive dataset. The model building unit is used to construct an initial model using an ensemble learning method, set the input and output variables of the initial model, and train the model using the comprehensive dataset to obtain a growth status assessment model. The input variables include canopy coverage, plant height, vegetation index, texture features, and environmental data. The output variables include a comprehensive growth status score, a health index, and a stress index. The indicator quantification unit is used to generate a spatial distribution map based on the comprehensive growth status score, and combine the health index and adversity index to obtain evaluation indicators for comprehensive assessment. The expression for the comprehensive growth status score is: in, To score the overall growth status, Canopy coverage, Plant height The vegetation index, For texture features, , , , The weights of each indicator; The expression for the health index is: in, For health index, , All are weighted parameters; The expression for the adversity index is: in, The adversity index. For soil moisture, For temperature, The optimal temperature for cotton growth. , All are weighted parameters; The growth prediction module is used to predict yield based on the growth status assessment model, introduce yield-related factors, and preset growth abnormality thresholds to provide early warning of growth abnormalities. The real-time monitoring and decision support module is used to utilize a cloud service platform for distributed data storage and management. Using WebGIS technology, it visualizes the comprehensive assessment results and growth prediction results of each cotton plot and automatically generates decision support based on the comprehensive assessment results and growth prediction results, and performs automated management based on the decision support. The data acquisition module, the data processing module, the image mining module, the comprehensive evaluation module, the growth prediction module, and the real-time monitoring and decision support module are interconnected.

2. The cotton field growth status assessment and prediction system based on UAV RGB images according to claim 1, characterized in that, The data acquisition module includes: The hardware configuration unit is used to select a drone and equip the selected drone with an RGB camera and a three-axis gimbal adapted to the RGB camera to achieve the camera's image stabilization performance. The flight path planning unit is used to automatically generate flight paths using UAV flight path planning software and to set the forward and lateral overlap of images. The parameter setting unit is used to set the drone's flight altitude and resolution according to the cotton growth stage, and then collect data to obtain initial image data.

3. The cotton field growth status assessment and prediction system based on UAV RGB images according to claim 2, characterized in that, The data processing module includes: The denoising and distortion correction unit is used to denoise the initial image data using automatic batch processing tools and denoising algorithms, and to correct the distortion of the image edges using camera calibration parameters. The orthophoto stitching unit is used to perform feature point matching and image stitching on the overlapping images in the initial image data using aerial survey processing software, and output an orthophoto with geographic coordinate system information. The illumination correction unit is used to deploy a grayscale plate in the flight area of ​​the UAV, record standard reflectance values, perform color correction on the initial image data based on the standard reflectance values, and integrate a radiometric correction algorithm in the aerial survey processing software to standardize the image brightness and image color.

4. The cotton field growth status assessment and prediction system based on UAV RGB images according to claim 3, characterized in that, The image mining module includes: The data annotation unit is used to perform pixel-level annotation based on the standard image data using annotation tools, generate segmentation labels for cotton plants and soil background, and annotate the bounding boxes of individual cotton plants and pest and disease areas. An image segmentation unit is used to select a convolutional neural network, introduce an attention mechanism, a multi-scale feature extraction module, and a depthwise separable convolution into the selected convolutional neural network to obtain a first initial model, and then train the first initial model using a deep learning framework to obtain a segmentation model. The segmentation model is used to segment the standard image data and output a binary image; wherein, the binary image includes the overall distribution information of the cotton canopy. The target detection unit is used to combine YOLOv5 and multi-task learning to obtain a target detection model, use the target detection model to detect the binary image, and output location information; wherein, the location information includes the location of a single cotton plant or a pest and disease area; The result fusion unit is used to combine the binary image and the location information to obtain cotton target information; The profile feature and growth index extraction unit is used to extract the canopy coverage, plant height, color features and vegetation index, as well as texture and spatial structure features of the cotton target information, to obtain the profile features and growth index of the cotton target information.

5. The cotton field growth status assessment and prediction system based on UAV RGB images according to claim 4, characterized in that, The profile feature and growth index extraction unit includes: The canopy coverage calculation subunit is used to count the percentage of cotton canopy pixels in the binary image to obtain the canopy coverage, and to assess whether the growth process is normal based on the canopy coverage and the cotton growth stage. The plant height estimation subunit is used to estimate the plant height by combining the orthophoto and the digital surface model, and output a spatial distribution map of the plant height to identify areas of abnormal growth. The texture and spatial structure feature extraction subunit is used to extract texture features based on the standard image data using the gray-level co-occurrence matrix, and to calculate the population density and spacing of cotton plants to obtain spatial structure features. The color feature and vegetation index calculation subunit is used to extract the color features of the RGB band in the standard image data, combine the texture features and the color features to design the RGB vegetation index, and then assess the health status of cotton and identify pest and disease areas based on the spatial distribution of the RGB vegetation index.

6. The cotton field growth status assessment and prediction system based on UAV RGB images according to claim 5, characterized in that, The growth prediction module includes: The yield prediction unit is used to fuse the canopy coverage, plant height, vegetation index, texture features, environmental data, and yield-related factors to obtain a yield prediction dataset. The yield prediction dataset is then used to train a model to obtain a yield prediction model. This model outputs predicted yield values ​​per plant and per unit area. The yield-related factors include plant density and fluff quantity. The yield prediction model incorporates environmental correction factors to adjust the output values. The growth anomaly unit is used to set a comprehensive growth status score threshold, a health index threshold, and an adversity index threshold based on the output value of the growth status assessment model. Then, based on the set thresholds, it determines whether the cotton field is an abnormal area. If so, it automatically generates an early warning message.

7. The cotton field growth status assessment and prediction system based on UAV RGB images according to claim 6, characterized in that: The expression for the predicted yield per plant is: in, This is the predicted yield per plant. For the amount of fluff produced, This refers to the weight of a single lint. The expression for the predicted yield per unit area is: in, This is the predicted yield per unit area. Plant density, To achieve optimal soil moisture, For optimal growth temperature, , This is the environmental correction factor.

8. The cotton field growth status assessment and prediction system based on UAV RGB images according to claim 7, characterized in that, The real-time monitoring and decision support module includes: The data storage unit is used to store structured data using relational databases based on cloud service platforms, manage unstructured data using object storage, and access data in real time through API interfaces. The monitoring and visualization unit is used to develop a responsive interface that supports multi-terminal access using a front-end framework, and to use WebGIS technology to display the processing results of the image mining module, the comprehensive evaluation module and the growth prediction module in real time on the interface through map overlay, plot view and heat map. The decision support unit is used to generate fertilization and irrigation suggestions, pest and disease control and supervision suggestions, climate change trend analysis and pest and disease pattern analysis based on the processing results of the image mining module, the comprehensive evaluation module and the growth prediction module.

Images