A deep learning-based citrus big data real-time analysis processing device

Abstract

The application discloses the technical fields of agricultural big data analysis, and discloses a citrus big data real-time analysis processing device based on deep learning, which comprises a collection module for collecting plant information, a pretreatment module, an identification and processing module for identifying and extracting abnormal image features, a three-dimensional model reconstruction module, a big data comparison module for inputting the three-dimensional model into a database for comparison, comprehensively evaluating the disease and pest risk grade and the diffusion trend, generating specific prevention and control strategies and executing basic prevention and control, a visual report module and a tracking feedback module. The application is novel and effective, and through shooting basic image information of citrus plants, images of different light transmission angles and fluorescence intensity images, combining convolutional neural networks to extract and model abnormal features in the images, comparing and evaluating accurate disease and pest species and stages in the database, accurate prediction of citrus diseases and pests is realized, basic prevention and control is automatically executed, the demand for manpower is reduced, and the prevention and control efficiency is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of agricultural big data analysis, and specifically relates to a deep learning-based real-time analysis and processing device for citrus big data. BACKGROUND

[0002] Plant diseases and insect pests are the biggest threat to the quality of agricultural planting production, and the prediction and control of plant diseases and insect pests are important directions for improving production quality and promoting agricultural economic growth. Citrus is a tree of the genus Citrus of the Rutaceae family, which is widely planted in Asia and the Americas, and its fruits are rich in nutrients such as vitamin C, fiber, minerals and antioxidants, and are one of the largest agricultural products traded in the world. However, during the growth and development of citrus to fruiting, it may be threatened by more than 100 diseases and insect pests, among which the representative diseases and insect pests are Huanglongbing, scab, sooty mold, etc. Lightly, it causes poor fruit development, affects quality, and heavily, it causes bad fruit and tree damage, causing significant economic losses.

[0003] Traditional citrus tree disease and insect pest monitoring relies on manual experience for diagnosis, but on the one hand, manual diagnosis is likely to be wrong, and on the other hand, the extensive citrus plantations rely on manual diagnosis, which consumes too much manpower. In addition to manual diagnosis methods, molecular biology-based detection and fluorescence biology-based detection have been developed, but these detection methods are difficult to achieve in situ detection of samples, and the number of samples is also limited by cost. Although the symptoms of citrus plant diseases and insect pests often appear on the leaf surface and fruit surface, some diseases and insect pests will cover the citrus leaf surface, causing phenomena similar to natural dust accumulation and water droplet, making it difficult to simply determine the type of citrus disease and insect pest through the plant surface image.

[0004] Therefore, it is necessary to propose a device that collects basic image information and light-supplemented leaf transmittance images, combines deep learning algorithms and big data technology, realizes comprehensive diagnosis of citrus diseases and insect pests and symptoms through three-dimensional modeling of citrus plants, generates prevention and control strategies and visual reports, improves the accuracy of citrus disease and insect pest prediction, reduces manpower consumption, and prevents the expansion of diseases and insect pests. SUMMARY

[0005] To solve the above problems, the purpose of the present application is to provide a deep learning-based real-time analysis and processing device for citrus big data, which captures basic image information of citrus plants, uses a light supplementing device to supplement light to plant leaves at night, captures images and fluorescence intensity images at different light transmission angles, identifies and extracts abnormal features in the images using a convolutional neural network, compares and evaluates accurate disease and insect pest species and stages in the database after surface reconstruction of the three-dimensional model, realizes accurate prediction of citrus diseases and insect pests, and automatically performs basic prevention and control, reduces the demand for manpower, and improves the efficiency of prevention and control.

[0006] In order to achieve the above object, the technical scheme of the present application is as follows: A deep learning-based citrus big data real-time analysis and processing device, comprising a collection module, a preprocessing module, an identification and processing module, a three-dimensional model reconstruction module, a big data comparison module, a visual report module, and a tracking feedback module;

[0007] The collection module is used to collect plantation environment information, plant image information, and plant physiological information, and comprises an environment sensor, a plant image collection device, and a plant physiological sensor.

[0008] The preprocessing module is used to arrange the information collected by the collection module according to time sequence, and to preprocess the plant image information and the plant physiological information based on the plantation environment information under the time sequence, including image scaling, normalization, and light correction.

[0009] The identification and processing module is used to identify and extract features in the plant image information and the plant physiological information by using a convolutional neural network trained by the preprocessed image, including abnormal color blocks on the surface of citrus plants, leaf shapes, leaf flatness, abnormal color blocks of citrus fruits, leaf transmittance image changes, and fluorescence intensity changes.

[0010] The three-dimensional model reconstruction module is used to reconstruct a three-dimensional model of a citrus plant using the preprocessed plant image information, and to reconstruct the surface of the model based on the feature extraction result of the convolutional neural network, thereby obtaining a three-dimensional model of the citrus plant containing key extracted features.

[0011] The big data comparison module is used to input the reconstructed three-dimensional model containing key extracted features in the plant image information and the plant physiological information into a database in a cloud server for comparison, and to comprehensively evaluate the disease and pest risk level and the spread trend of the citrus plant based on the plantation environment information and the statistical information of the disease and pest organisms in the database, thereby generating a specific prevention and control strategy based on the disease and pest risk level and executing basic prevention and control.

[0012] The visual report module is used to arrange and output the three-dimensional model reconstructed based on the feature extraction result of the convolutional neural network, the disease and pest risk level evaluation and the spread trend, and the collected plantation environment information as a visual three-dimensional model report.

[0013] The tracking feedback module is used to track the plant performing the basic prevention and control according to the content of the prevention and control strategy, and to feed back the tracked plant image information and plant physiological information to the identification and processing module for analysis and modeling at the corresponding time node, and to output a tracking report.

[0014] The principle of the basic scheme is that the device collects and analyzes image information and physiological information of citrus plants, especially multi-angle leaf transmittance images and fluorescence intensity images after night light supplement, combines deep learning algorithms and big data technology, realizes accurate prediction and prevention of citrus diseases and pests, and improves the accuracy of disease and pest prediction, reduces labor consumption, and effectively prevents the spread of diseases and pests.

[0015] The beneficial effects of the basic scheme are: 1. By integrating advanced deep learning algorithms and big data technology, the device can efficiently extract key features from massive image information and physiological data, especially abnormal features of each part in the reconstructed three-dimensional model, to realize accurate prediction of citrus diseases and pests. Compared with traditional manual diagnosis methods, this scheme greatly reduces the misdiagnosis rate and improves the accuracy and timeliness of diagnosis. This not only helps growers to take timely measures, but also effectively prevents the further spread of diseases and pests, ensuring the yield and quality of citrus.

[0016] 2. The automatic and intelligent monitoring and diagnosis process reduces the need for human intervention, thereby significantly reducing labor costs. Growers do not need to frequently patrol the orchard, but can obtain real-time health information of citrus plants through the device, and automatically perform basic disease and pest control treatment. This not only reduces the labor intensity of growers, but also enables them to devote more energy to the daily management and optimization of the orchard.

[0017] 3. Combined with three-dimensional model reconstruction and big data comparison technology, the device is not limited to judging symptoms on the surface of citrus leaves and fruits, but can also identify some early symptoms of plant diseases and pests through abnormalities in each part of the three-dimensional model, and provide scientific and specific control strategies for growers. These strategies are based on accurate prediction and risk assessment of diseases and pests, and have high pertinence and effectiveness. By implementing these strategies, growers can more accurately control diseases and pests, reduce the amount of pesticides used, and reduce environmental pollution.

[0018] 4. The visualization report module of the device can present complex disease and pest information in an intuitive and easy-to-understand manner, providing strong support for the informatization management of the orchard. Growers can quickly understand the overall health status of the orchard and the distribution of diseases and pests in specific areas through these reports. At the same time, the tracking feedback module can also monitor the control effect in real time to provide data support for subsequent orchard management.

[0019] Further, the environmental sensors in the acquisition module include temperature and humidity sensors, soil property sensors, light sensors, rainfall sensors, and wind speed and direction sensors.

[0020] The planting environment information collected by the collection module includes preset planting terrain information and weather information, and real-time monitored air temperature, air humidity, soil humidity, soil acidity and alkalinity, light intensity, light direction, rainfall, water accumulation, wind direction and wind speed information.

[0021] The beneficial effects of the basic scheme are: 1. Environmental information is one of the important factors affecting the occurrence of pests and diseases. By monitoring the environmental factors such as temperature, humidity and light in the plantation in real time, we can introduce environmental information into the preprocessing module to normalize and correct the plant image information and plant physiological information used for identification more scientifically, so as to obtain more accurate prediction of the occurrence time and intensity of pests and diseases, and provide timely warning information for the grower to help them take measures in advance for prevention and control.

[0022] 2. Understanding the environmental information in the plantation helps the grower to adjust irrigation, fertilization, shading and other management measures according to the growth needs of the plants, to provide a more suitable growing environment for citrus plants. This not only helps to improve the yield and quality of citrus, but also reduces the occurrence of pests and diseases and reduces the prevention and control cost.

[0023] 3. Combined with the real-time monitoring of environmental information and the prediction results of pests and diseases, we can provide more scientific and specific prevention and control strategies for the grower. These strategies not only consider the types and damage of pests and diseases, but also fully consider the influence of environmental factors, so as to improve the pertinence and effectiveness of prevention and control.

[0024] Further, the plant image collection device in the collection module includes a plurality of high-definition cameras, a camera carried by a drone, and a bottom light supplementing device;

[0025] The plant image information collected by the collection module includes high-definition image information of the plants in different directions, leaf transmittance image information after multi-source angle light supplementing by the bottom light supplementing device at night, and image information of the roots, branches, crowns and treetops of citrus plants.

[0026] The beneficial effects of the basic scheme are: 1. High-definition cameras and cameras carried by drones can capture images of plants from multiple angles to ensure comprehensive monitoring of pests and diseases, and provide solid data support for the reconstruction of three-dimensional plant models. The leaf transmittance image information at night can reveal the pests and diseases inside the leaves, and exclude the identification errors caused by dirt covering the surface of the leaves. The image information of the roots, branches, crowns and treetops helps to identify the characteristics of pests and diseases in different parts. These information provide strong support for accurate identification of pests and diseases, and help the grower to take measures for prevention and control in time.

[0027] 2、By collecting high-definition image information in different directions, we can monitor the growth status of citrus plants in a comprehensive and real-time manner. This helps growers understand the growth trend, nutritional status, and morphological changes of the leaves, so as to adjust the management measures and promote the healthy growth of the plants. Combined with image information and environmental data, we can provide more specific and targeted prevention and control strategies for growers. These strategies not only consider the types and damage of pests and diseases, but also consider the growth status and nutritional status of the plants, thereby improving the accuracy and effectiveness of prevention and control.

[0028] 3、Through the data obtained by the image acquisition device, we can also use machine learning algorithms to intelligently analyze the growth environment of the orchard, realizing the intelligent management of the orchard. For example, we can establish an orchard growth model to simulate and optimize the management measures such as irrigation and fertilization, thereby improving the management efficiency and resource utilization efficiency of the orchard.

[0029] Further, the plant physiological sensor in the acquisition module includes a spectrum sensor and a chlorophyll excitation light source in the bottom light supplementing device;

[0030] The plant physiological information collected by the acquisition module is the chlorophyll fluorescence image information obtained by exciting the fluorescence of the citrus leaf at night using the chlorophyll excitation light source and collected by the spectrum sensor.

[0031] The beneficial effects of the basic scheme are: 1、The chlorophyll fluorescence image information can reflect the photosynthetic efficiency, nutritional status, and physiological response under stress of the citrus leaf. Through the data collected by the spectrum sensor, we can accurately evaluate the physiological indicators such as chlorophyll content and photosynthetic rate of the leaf, thereby understanding the growth potential and health status of the plant.

[0032] 2、The chlorophyll fluorescence image information can also reveal the pest and disease situation inside the leaf. Through the characteristics (such as different fluorescence shielding and fluorescence loss) displayed by the chlorophyll fluorescence of the leaf, we can also assist in determining whether the image information collection is affected by dirt. By comparing the fluorescence images of normal leaves and leaves affected by pests and diseases, we can more accurately identify the types and damage of pests and diseases, thereby developing more effective prevention and control strategies.

[0033] Further, the normalization processing of the preprocessing module is to divide each pixel value in the plant image information and the plant physiological information by the standard deviation of the pixel value, and perform standard deviation normalization.

[0034] The beneficial effects of the basic scheme are: 1. Normalization processing converts data of different scales and distributions to a unified scale, thereby simplifying the complexity of data processing. In subsequent image processing or data analysis processes, since all data are within the same scale range, the computational efficiency and convergence speed of the algorithm will be improved. The normalized data reduces the sensitivity of the model to the scale change of the input data, making the model more stable when facing data of different scales. By improving the consistency and comparability of data, the model can better learn the internal laws and characteristics of the data, thereby improving its generalization ability.

[0035] 2. In machine learning and deep learning models, normalization processing helps to speed up the convergence speed of the gradient descent algorithm, so that the model can reach the optimal solution faster. Normalization can also reduce the problem of gradient vanishing or explosion, especially in neural networks, which helps the model to better learn the complex relationships between features. 3. In image processing and plant physiological information analysis, normalized data can more accurately reflect the growth status and physiological characteristics of plants, thereby providing a more reliable basis for subsequent prediction and analysis, and helping to reduce the interference of outliers on model training and prediction results.

[0037] Further, the light correction processing of the preprocessing module is based on the light direction and light intensity information in the planting yard environment information, and the plant image information collected at the corresponding time is corrected in brightness and light.

[0038] The beneficial effects of the basic scheme are: 1. By comparing the light intensity in the planting yard environment information, the brightness of the image can be adjusted to ensure that the image maintains consistent brightness levels under different lighting conditions. Brightness correction helps to eliminate image blurring and detail loss caused by insufficient or excessive exposure, making the image clearer. According to the light direction information, the image can be rotated or locally adjusted in brightness to eliminate image distortion caused by light inclination. Light direction correction helps to restore the true shape and outline of the image, improving the accuracy of the image.

[0039] 2. After light correction, the features of the target object in the image (such as edges, textures, etc.) are more prominent, which is beneficial for subsequent feature extraction and recognition work. Accurate feature extraction is the basis for image analysis, target detection and classification tasks, which helps to improve the accuracy of analysis. Light correction can eliminate image color distortion and shadow interference caused by changes in lighting, thereby reducing the possibility of misjudgment. In plant disease and pest detection, light correction helps to improve the accuracy and reliability of detection.

[0040] 3、Light corrected image data is more consistent and stable, which is conducive to improving the adaptability and robustness of subsequent depth algorithms. The fusion of light corrected image data and environmental data (such as temperature, humidity, etc.) is more smooth, which is conducive to building a more complete and accurate plantation information model.

[0041] Further, the big data comparison module includes a database, a risk assessment unit, a strategy generation unit, and a prevention and control execution unit;

[0042] The database includes a cloud server with a decentralized architecture, which is used to compare the three-dimensional plant model reconstructed by the extracted feature surface to obtain disease and pest possibility prediction data;

[0043] The risk assessment unit is used to receive disease and pest possibility prediction data and compare plant model differences in different positions in the plantation to exclude individual accidents, and comprehensively assess the risk level and spread trend of the current plant according to disease and pest type, disease and pest onset period, and disease and pest biological statistics;

[0044] The strategy generation unit is used to generate specific prevention and control strategies according to the plant risk level obtained by evaluation, including prevention and control methods, prevention and control implementation time, prevention and control duration, and prevention and control protection treatment;

[0045] The prevention and control execution unit is used to automatically execute basic disease and pest control according to the generated prevention and control strategy, including replacing the camera module of the unmanned aerial vehicle with an agricultural spraying module to spray pesticides on the plant surface for prevention and control, and replacing the unmanned aerial vehicle with a moth lamp module to achieve patrol type pest control.

[0046] The beneficial effects of the basic scheme are: 1. By storing and comparing a large amount of three-dimensional plant model data through a cloud server with a decentralized architecture, the accuracy of disease and pest prediction is improved with the support of big data. Three-dimensional model reconstruction technology can reflect the growth state and disease and pest characteristics of plants from multiple angles and levels, improving the accuracy of identification. Key features are extracted from the three-dimensional plant model and compared with disease and pest characteristics in the database to quickly identify disease and pest types and infection levels. This method based on big data and three-dimensional models is more accurate and efficient than traditional manual identification.

[0047] 2. Consider disease and pest type, onset period, and biological statistics to analyze differences in plant models at different positions and exclude individual accidents. Through comparative analysis, the risk level and spread trend of the current plant are scientifically evaluated to provide a reliable basis for developing prevention and control strategies. The risk assessment results can reflect the actual situation and potential threat of disease and pests, making prevention and control decisions more accurate and targeted. This helps to reduce the waste of prevention and control resources and improve prevention and control effectiveness.

[0048] 3. According to the risk assessment results, specific prevention and control strategies are automatically generated, including prevention and control methods, implementation time, duration and protection measures. Intelligent strategy generation can reduce human intervention and improve the efficiency and accuracy of decision-making. Unmanned aerial vehicles and other automated equipment can automatically execute pest control tasks according to the generated prevention and control strategies. Different modules (such as agricultural spraying modules and insect light modules) carried by unmanned aerial vehicles can achieve flexible switching and efficient execution of various prevention and control methods. Automated and intelligent prevention and control execution methods can greatly shorten the prevention and control cycle and improve prevention and control efficiency.

[0049] Further, the specific strategy generated by the strategy generation unit includes,

[0050] The prevention and control method includes spraying pesticides, building diseased branches, fertilizer control, burning and burying diseased plants, loosening soil and draining water, and bagging young fruits.

[0051] The prevention and control implementation time refers to the best implementation time of the prevention and control measures according to the assessed risk level.

[0052] The prevention and control duration refers to the estimated time required to suppress pests and diseases after taking prevention and control measures according to the assessed risk level.

[0053] The prevention and control protection refers to the protection treatment for normal plants to prevent and control the spread according to different prevention and control methods.

[0054] The beneficial effects of the basic scheme are: 1. The selection of multiple prevention and control methods such as spraying pesticides, building diseased branches, fertilizer control, burning and burying diseased plants, loosening soil and draining water, and bagging young fruits allows the prevention and control measures to be flexibly adjusted according to different types of pests and diseases, infection levels, and plant growth states. This diversity not only improves the flexibility of prevention and control, but also enhances the effectiveness and effectiveness of prevention and control. According to the assessed risk level, the best implementation time is determined to ensure that the prevention and control measures are in place at the critical period of pest and disease development, effectively controlling the development of the disease. Precise implementation time helps to improve prevention and control efficiency and reduce waste of prevention and control resources.

[0055] 2. According to the assessed risk level, the estimated time required to suppress pests and diseases can reasonably arrange the duration and frequency of prevention and control. This rationality not only avoids the waste of resources caused by excessive prevention and control, but also ensures the durability of the prevention and control effect. According to different prevention and control methods, targeted protection treatment can protect normal plants from pests and diseases. Targeted protection treatment helps to reduce the spread of pests and diseases and reduce prevention and control costs.

[0056] Further, the visual three-dimensional model report of the visual report module is based on the preset plantation terrain information, combines the three-dimensional model reconstructed by the feature extraction result of the convolutional neural network, the pest and disease risk level evaluation and the diffusion trend, obtains a visual plantation scene model, and renders the diffusion trend on the model surface in the form of a time contour graph.

[0057] The beneficial effects of the basic scheme are: 1. The visual report intuitively displays the scene of the plantation in the form of a three-dimensional model, including the terrain, plant distribution and diffusion trend of pests and diseases. This intuitiveness enables decision makers to quickly understand the overall situation of the plantation, reducing the time cost of information interpretation. By rendering the time contour graph on the surface of the model, the diffusion trend and risk level of pests and diseases are clearly displayed. Decision makers can quickly judge the severity and diffusion speed of pests and diseases based on this information, making more accurate decisions.

[0058] 2. The visual report can accurately locate the areas where pests and diseases occur and the diffusion path, so that the prevention and control measures can be implemented in a targeted manner. This helps to reduce the waste of prevention and control resources and improve resource utilization efficiency. The visual report module can update the diffusion trend and risk level of pests and diseases in real time, providing a basis for dynamic monitoring for decision makers. Through continuous monitoring, decision makers can adjust the prevention and control strategy in a timely manner to ensure the prevention and control effect.

[0059] 3. The report generated by the visual report module is easy to understand and share, and can be easily communicated to plantation managers, technical personnel and decision makers and other different roles. This helps to enhance communication and cooperation among teams to jointly address the challenges of pests and diseases. The visual report provides intuitive and comprehensive information support for decision makers, helping to form more scientific and reasonable prevention and control strategies. By optimizing the prevention and control strategy, the impact of pests and diseases on the plantation can be reduced, and the overall economic benefit can be improved.

[0060] Further, the bottom light supplementing device comprises a reflection ring, a rotating motor is fixedly connected to the top wall of the reflection ring, the output shaft of the rotating motor is vertically upward and coaxially fixedly connected with a transmission gear, the transmission gear is engaged with the outer periphery of a gear ring, the top wall of the gear ring is coaxially fixedly connected with a lamp ring, the lamp ring is irregular in axial direction, the reflection ring is fixedly connected with the chlorophyll excitation light source in outer periphery, the reflection ring is fixedly sleeved with the middle part of the citrus plant trunk, and the gear ring and the lamp ring are both slidingly sleeved with the middle part of the citrus plant trunk.

[0061] The beneficial effects of the basic scheme are: 1. The lamp ring is irregular in axial direction, and can provide angle conversion illumination for different parts of the citrus canopy when rotating. This design helps to increase the unevenness of the illumination, improve the degree of light transmission change in the collected plant leaf light transmission image, and accurately exclude the error caused by dirt covering.

[0062] 2, the reflection ring can reflect the light emitted by the lamp ring and the chlorophyll excitation light source, further enhancing the light supplementing effect. The chlorophyll excitation light source can excite the chlorophyll fluorescence in the citrus leaf, which helps to more accurately detect the health status of the leaf. And the multi-angle image of chlorophyll fluorescence can extract and identify whether the leaf produces bad spots, which plays an auxiliary role in accurate identification of diseases and pests. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 It is a schematic diagram of the deep learning-based citrus big data real-time analysis processing device in the embodiment of the application.

[0064] Figure 2 It is a schematic diagram of the acquisition module of the deep learning-based citrus big data real-time analysis processing device in the embodiment of the application.

[0065] Figure 3 It is a schematic diagram of the deep learning-based citrus big data real-time analysis processing device in the embodiment of the application.

[0066] Figure 4 It is an axonometric view of the bottom light supplementing equipment of the deep learning-based citrus big data real-time analysis processing device in the embodiment of the application.

[0067] The reference signs in the drawings of the specification include: 1, plant trunk; 2, reflection ring; 3, rotating motor; 4, transmission gear; 5, gear ring; 6, lamp ring; 7, chlorophyll excitation light source; 8, reflecting mirror surface. DETAILED DESCRIPTION

[0068] The following will be further described in detail through specific embodiments:

[0069] Embodiment 1:

[0070] Basically as shown in the accompanying drawings: a deep learning-based citrus big data real-time analysis processing device, comprising an acquisition module and a preprocessing module. Figure 1 and Figure 2 Basically as shown in the accompanying drawings: a deep learning-based citrus big data real-time analysis processing device, comprising an acquisition module and a preprocessing module.

[0071] The collection module is used for collecting plantation environment information, plant image information and plant physiological information, and includes an environment sensor, a plant image collection device and a plant physiological sensor; the environment sensor includes a temperature and humidity sensor, a soil property sensor, a light sensor, a rainfall sensor and a wind speed and direction sensor; the plantation environment information includes preset plantation terrain information and weather information, and real-time monitored air temperature, air humidity, soil humidity, soil acidity and alkalinity, light intensity, light direction, rainfall, water accumulation, wind direction and wind speed information; the plant image collection device includes a plurality of high-definition cameras, a camera carried by a drone and a bottom light supplementing device; the plant image information includes high-definition plant image information in different directions, leaf transmittance image information after multi-source angle light supplementing by the bottom light supplementing device at night, and image information of parts of a citrus plant, such as roots, branches, a crown and a top; the plant physiological sensor includes a spectrum sensor and a chlorophyll excitation light source in the bottom light supplementing device; the plant physiological information is chlorophyll fluorescence image information obtained by exciting citrus leaf fluorescence by the chlorophyll excitation light source at night and collected by the spectrum sensor.

[0072] The preprocessing module is used for arranging information collected by the collection module according to a time sequence, and pre-processing plant image information and plant physiological information based on plantation environment information under the time sequence, including image scaling, normalization and light correction; the normalization processing is to divide each pixel value in the plant image information and the plant physiological information by a standard deviation of pixel values, for standard deviation normalization; the light correction processing is to correct brightness and light of plant image information collected at a corresponding time based on light direction and light intensity information in the plantation environment information arranged according to the time sequence.

[0073] The specific implementation process is as follows: the various plantation environment information, plant image information and plant physiological information collected in the collection module mainly provide perfect and effective data support for subsequent image preprocessing, image neural network learning, three-dimensional model reconstruction and big data comparison, especially the light intensity and light direction energy are used for normalization and light correction of the plant image information and the plant physiological information in the preprocessing module, to reduce the influence of reflected light and brightness difference on feature recognition and extraction of the subsequent convolutional neural network.

[0074] As shown in Figure 2 , the plant image information photographed by the high-definition camera and the camera carried by the drone is an important basis for plant three-dimensional model reconstruction, multi-angle views can provide depth information for three-dimensional model reconstruction, and image clarity directly affects the fineness of the three-dimensional model reconstruction grid.

[0075] Since naturally occurring dust and other stains may adhere to the surface of plant leaves, direct image recognition by neural networks is highly likely to cause misjudgments. For example, citrus sooty mold forms black mold spots on the surface of plant leaves, which then spread and cover the entire leaf. Its appearance is similar to the dust that naturally accumulates on the leaf surface. Therefore, it is necessary to use supplemental lighting equipment to illuminate the plant from the bottom and take images of the plant leaves at night from different light source angles. The neural network can use these images to determine whether the black mold spots are only covering the leaf surface or growing from within the leaf, thus eliminating misjudgments caused by accidental external stains.

[0076] In the process of big data comparison, it is necessary not only to compare the identification and extraction features of each part in the surface reconstruction data of the 3D reconstruction model, which are the direct basis for judging the occurrence period and type of citrus plant diseases and pests, but also to use other orchard environmental information to indirectly corroborate the incidence probability of diseases and pests, enrich the data foundation for comparison, and improve the efficiency and accuracy of comparison results. For example, citrus red spider mite infestation occurs in groups of adult mites, nymphs, and larvae on citrus leaves, tender branches, and fruits. They use their mouthparts to pierce the epidermis of leaves, tender shoots, and fruits and suck sap. The optimal temperature for the reproduction and development of red spider mites is 20-30℃, and the peak season for damage is spring and autumn. Orchard temperature information can corroborate whether the current period is the peak period of pest occurrence. If so, the probability weight of this judgment is increased. Citrus sunburn, a non-biological disease, is caused by excessive sunlight, which leads to leaf drying, yellowing and browning of fruit tops, and stunted development. The identification and extraction features of leaf and fruit images, combined with the light intensity information in the orchard environmental information, can provide sufficient basis for comparison and obtain sufficiently accurate probability results, providing accurate data support for further generating specific control strategies.

[0077] Furthermore, collecting multi-angle chlorophyll fluorescence images at night can help determine whether the fluorescence is blocked by stains or due to loss of fluorescent tissue, thus helping to eliminate interference from stains in image feature extraction. In addition, the intensity and spectral changes of chlorophyll fluorescence can reflect the growth status of citrus plants and can serve as a basis for judging whether citrus plants are infected by pests or diseases. For example, in plants infected with Huanglongbing (HLB), the minimum chlorophyll fluorescence value increases and the proportion of fluorescence quenching rises. This is because the photosystem II reaction centers of the chloroplasts in the leaves are damaged, the maximum photon efficiency decreases, and the photochemical reaction capacity of the infected leaves is reduced.

[0078] Example 2:

[0079] The difference from the above embodiments is that, as shown in the appendix Figure 3 As shown, the deep learning-based real-time analysis and processing device for citrus big data also includes an identification and processing module, a 3D model reconstruction module, a big data comparison module, a visualization and reporting module, and a tracking and feedback module.

[0080] The recognition processing module is configured to recognize and extract features in the plant image information and the plant physiological information by using a convolutional neural network trained by the preprocessed images, including abnormal color blocks on the surface of the citrus plants, leaf shapes, leaf flatness, abnormal color blocks of the citrus fruits, leaf transmittance image changes, and fluorescence intensity changes.

[0081] The three-dimensional model reconstruction module is configured to reconstruct a three-dimensional model of the citrus plants from multiple views using the preprocessed plant image information, and reconstruct a surface of the model based on the feature extraction results of the convolutional neural network, to obtain a three-dimensional model of the citrus plants containing the key extracted features.

[0082] The big data comparison module is configured to input the reconstructed three-dimensional model containing the key extracted features in the plant image information and the plant physiological information into a database in a cloud server for comparison, and comprehensively evaluate a pest and disease risk level and a spreading trend of the citrus plants based on planting yard environment information and statistical information of the pests and diseases in the database, to generate a specific prevention and control strategy based on the pest and disease risk level and execute basic prevention and control. Figure 3 As shown in the figure, the big data comparison module includes a database, a risk evaluation unit, a strategy generation unit, and a prevention and control execution unit. The database includes a cloud server with a decentralized architecture, and is configured to compare the three-dimensional model of the plants reconstructed based on the extracted features to obtain pest and disease possibility prediction data. The risk evaluation unit is configured to receive the pest and disease possibility prediction data, compare differences in the plant models at different positions in the planting yard, exclude individual accidents, and comprehensively evaluate a risk level and a spreading trend of the current plants based on the type of the pests and diseases, the incidence period of the pests and diseases, and statistical data of the pests and diseases. The strategy generation unit is configured to generate a specific prevention and control strategy based on the evaluated risk level of the plants, including a prevention and control method, a prevention and control implementation time, a prevention and control duration, and a prevention and control protection treatment. The prevention and control execution unit is configured to automatically execute basic pest and disease prevention and control based on the generated prevention and control strategy, including replacing a camera module of a drone with an agricultural spraying module to spray pesticides on the surface of the plants, and replacing the drone with a moth lamp module to achieve a patrol type pest killing. In the specific strategy generated by the strategy generation unit, the prevention and control method includes spraying pesticides, pruning diseased branches, controlling fertilizers, burning and burying diseased plants, loosening soil and draining water, and bagging young fruits. The prevention and control implementation time refers to the best implementation time of the prevention and control measures based on the evaluated risk level. The prevention and control duration refers to the predicted time required to inhibit the pests and diseases after taking the prevention and control measures based on the evaluated risk level. The prevention and control protection treatment refers to the protection treatment for normal plants to prevent the spread according to different prevention and control methods.

[0083] The visual report module is used for combining the three-dimensional model reconstructed by the feature extraction result of the convolutional neural network, the disease and pest risk level evaluation and the diffusion trend, and the collected planting garden environment information, and outputting a visual three-dimensional model report. The visual three-dimensional model report is based on the preset planting garden terrain information, combines the three-dimensional model reconstructed by the feature extraction result of the convolutional neural network, the disease and pest risk level evaluation and the diffusion trend, obtains a visual planting garden scene model, and renders the diffusion trend on the model surface in the form of a time contour line.

[0084] The tracking feedback module is used for tracking the plant images and plant physiological information of the plants performing the basic prevention according to the prevention strategy content, feeding back the tracked plant images and plant physiological information to the identification processing module for analysis and modeling at corresponding time nodes, and outputting a tracking report.

[0085] The specific implementation process is as follows: after the plant image information is preprocessed, the trained convolutional neural network performs feature recognition and extraction on the plant image information and the plant physiological information, mainly recognizing and extracting the position, color and shape changes of the leaves that are prone to show symptoms after the citrus is infected with diseases and pests, and also including the appearance of abnormal color blocks on the citrus fruits and the exclusion of surface stain interference through leaf transmittance image changes and fluorescence intensity changes. The leaves are the early symptom occurrence sites of diseases and pests, and the color and shape changes and occurrence positions thereof are important bases for judging the types and occurrence periods of subsequent comparison and judgment of diseases and pests. For example, the symptoms of Huanglongbing first occur at the top of summer and autumn shoots, the tender branches turn yellow, then spread downward, the leaves gradually become mottled, the leaf flesh becomes thick and hardened, the leaf surface loses luster, the leaf veins become enlarged, and the whole plant is infected after 1-2 years, so the convolutional neural network can first extract the position and color change of the tender branches that first get sick, as powerful data support for the early occurrence of Huanglongbing disease. Moreover, the convolutional neural network can continuously use new image data for continuous training to improve the accuracy of recognition and extraction.

[0086] The three-dimensional model reconstruction module reconstructs a three-dimensional model of the citrus plant according to multi-angle plant image information, and reconstructs the surface of the model in combination with the feature extraction result of the convolutional neural network, thereby improving the spatial proportion of feature extraction and disease and pest comparison and judgment, so that more comprehensive and perfect data information is input into the big data cloud server for comparison, the risk assessment unit in the big data comparison module can also compare the feature differences of plant models at different positions to exclude accidental misjudgment, obtain real and accurate results, and evaluate the risk level and diffusion trend of the current disease and pest symptoms in combination with the disease and pest biological statistical data in the big data. The diffusion trend also needs to be fitted and calculated by the risk assessment unit in combination with the disease and pest characteristic data in the big data and the planting garden environment information to obtain a diffusion trend graph.

[0087] After obtaining the risk level of pests and diseases, the strategy generation unit combines the occurrence period of pests and diseases to produce the most suitable prevention and control methods at the current stage, as well as the implementation time, duration of prevention and control, and protective treatment. It also controls the drone to flexibly switch prevention and control modules to achieve automatic prevention and control of diseased plants, reduce manpower requirements, and promptly suppress the occurrence and spread of pests and diseases.

[0088] The subsequent complete 3D model information and pest and disease prediction report, combined with orchard environmental information, are presented to the managers in the form of a model. It intuitively shows the location of diseased plants and the subsequent spread path and range, and reports scientific and comprehensive prevention and control strategies for managers to adopt. This realizes the prediction and intelligent monitoring of pests and diseases in citrus orchards, improves the accuracy of pest and disease prediction, reduces human intervention, and lowers planting costs.

[0089] Example 3:

[0090] The difference from the above embodiments is that, as shown in the appendix Figure 4 As shown: The bottom supplemental lighting device includes a reflective ring 2, the top wall of which is a reflective mirror 8, a rotating motor 3 welded to the top wall of the reflective ring 2, the output shaft of the rotating motor 3 is vertically upward and coaxially welded with a transmission gear 4, the transmission gear 4 meshes with the outer circumference of a gear ring 5, a lamp ring 6 is coaxially bonded to the top wall of the gear ring 5, the lamp ring 6 is irregularly shaped axially, the outer circumference of the reflective ring 2 is bolted to a chlorophyll excitation light source 7, the reflective ring 2 is bonded to the middle of the citrus tree trunk, and both the gear ring 5 and the lamp ring 6 are slidably sleeved with the middle of the citrus tree trunk 1.

[0091] The specific implementation process is as follows: Supplemental lighting for leaves is mainly used for acquiring multi-angle leaf translucency images and chlorophyll fluorescence images under low-light conditions. The reflector ring 2 reflects the light from the lamp ring 6, increasing the supplemental lighting intensity at the bottom of the leaves. During supplemental lighting, the rotating motor 3 drives the gear ring 5 to rotate via the transmission gear 4, which in turn drives the lamp ring 6 to rotate. During the irregular rotation of the lamp ring 6 along its axis, the direction and intensity of the light naturally change, achieving multi-angle supplemental lighting for the citrus plant leaves. Simultaneously, under low-light conditions, the chlorophyll excitation light source 7 can independently excite the citrus leaves. Combined with a spectral sensor, the chlorophyll fluorescence of the excited leaves is detected, reflecting the different layers of leaf blemishes and indicating the plant's growth and health status.

[0092] It is to be noted that, as used in this document, the term "indicates" a relationship between one entity or action and another entity or action, without necessarily requiring a direct physical or logical relationship between, or actual movement of, such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0093] The above description is only some embodiments of the present application, and the common knowledge of the specific structure and characteristics in the scheme is not described in detail. The ordinary skilled person in the art knows all the ordinary technical knowledge in the field of the present application before the application date or the priority date, can know all the prior art in the field, and has the ability to apply the conventional experimental means before the date. The ordinary skilled person in the art can perfect and implement the present scheme under the guidance of the present application, combined with their own ability. Some typical known structures or known methods should not be an obstacle for the ordinary skilled person in the art to implement the present application. It should be pointed out that, for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can be made, which should be considered as the protection scope of the present application. These will not affect the effect and practicality of the present application. The protection scope of the present application should be subject to the content of its claims, and the specific implementation mode in the specification can be used to explain the content of the claims.

Claims

1. A real-time analysis and processing device for citrus big data based on deep learning, characterized in that: It includes a data acquisition module, a preprocessing module, a recognition and processing module, a 3D model reconstruction module, a big data comparison module, a visualization and reporting module, and a tracking and feedback module; The data acquisition module is used to collect plantation environmental information, plant image information, and plant physiological information, including environmental sensors, plant image acquisition equipment, and plant physiological sensors. The preprocessing module is used to organize the information collected by the acquisition module according to the time series, and to preprocess the plant image information and plant physiological information based on the plantation environment information under the time series, including image scaling, normalization and light correction. The recognition and processing module is used to identify and extract features from plant image information and plant physiological information using a convolutional neural network trained on preprocessed images, including abnormal color patches on the surface of citrus plants, leaf shape, leaf flatness, abnormal color patches on citrus fruits, changes in leaf light transmission images, and changes in fluorescence intensity. The 3D model reconstruction module is used to reconstruct a 3D model of a citrus plant from multiple views using preprocessed plant image information. It combines the feature extraction results of the convolutional neural network to perform surface reconstruction of the model, resulting in a 3D model of a citrus plant containing key extracted features. The big data comparison module is used to input the reconstructed 3D model containing key extracted features from plant image information and plant physiological information into the database in the cloud server for comparison. It also combines plantation environmental information and pest and disease biostatistics information in the database to comprehensively assess the pest and disease risk level and spread trend of citrus plants, generate specific prevention and control strategies based on the pest and disease risk level, and implement basic prevention and control measures. The visualization report module is used to combine the 3D model reconstructed from the feature extraction results of the convolutional neural network, the pest and disease risk level assessment, and the spread trend with the collected plantation environmental information and output it as a visualized 3D model report. The tracking and feedback module is used to track the information of plants that have undergone basic prevention and control according to the prevention and control strategy, and to feed back the tracked plant image information and plant physiological information to the identification and processing module for analysis and modeling at the corresponding time nodes, and output a tracking report. The plant image acquisition equipment in the acquisition module includes several high-definition cameras, a camera mounted on a drone, and a bottom supplementary lighting device; The plant image information collected by the acquisition module includes high-definition images of the plant from different directions, leaf light transmission images after multi-source angle supplemental lighting using bottom supplemental lighting equipment at night, and image information of various parts of the citrus plant, including the roots, branches, crown, and treetops.

2. The deep learning-based real-time analysis and processing device for citrus big data according to claim 1, characterized in that: The environmental sensors in the data acquisition module include temperature and humidity sensors, soil property sensors, light sensors, rainfall sensors, and wind speed and direction sensors; The plantation environmental information collected by the data acquisition module includes preset plantation terrain information and weather information, as well as real-time monitoring of temperature, air humidity, soil moisture, soil pH, light intensity, light direction, rainfall, water accumulation, wind direction and wind speed.

3. The real-time analysis and processing device for citrus big data based on deep learning according to claim 1, characterized in that: The plant physiological sensors in the acquisition module include a spectral sensor and a chlorophyll excitation light source in the bottom supplemental lighting device. The plant physiological information acquired by the acquisition module is the chlorophyll fluorescence image information obtained by the spectral sensor at night when the chlorophyll excitation light source excites the fluorescence of citrus leaves.

4. The real-time analysis and processing device for citrus big data based on deep learning according to claim 1, characterized in that: The normalization process in the preprocessing module involves dividing each pixel value in the plant image information and plant physiological information by the standard deviation of the pixel value to perform standard deviation normalization.

5. The deep learning-based real-time analysis and processing device for citrus big data according to claim 1, characterized in that: The light correction process in the preprocessing module is based on the light direction and light intensity information in the plantation environment information organized in time sequence, and performs brightness and light correction on the plant image information collected at the corresponding time.

6. The real-time analysis and processing device for citrus big data based on deep learning according to claim 1, characterized in that: The big data comparison module includes a database, a risk assessment unit, a strategy generation unit, and a prevention and control execution unit. The database includes a distributed cloud server used to compare the three-dimensional plant models reconstructed from extracted feature surfaces to obtain pest and disease probability prediction data. The risk assessment unit is used to receive pest and disease probability prediction data and compare the differences in plant models in different locations in the plantation, eliminate individual randomness, and comprehensively assess the current risk level and spread trend of the plant based on pest and disease type, pest and disease onset time and pest and disease biostatistics. The strategy generation unit is used to generate specific control strategies based on the plant risk level obtained from the assessment, including control methods, control implementation time, control duration and control protection treatments. The prevention and control execution unit is used to automatically perform basic pest and disease control based on the generated prevention and control strategy. This includes replacing the camera module of the drone with an agricultural spraying module to spray pesticides on the plant surface, and replacing the drone with an insect-attracting lamp module to achieve patrol-style pest control.

7. The real-time analysis and processing device for citrus big data based on deep learning according to claim 6, characterized in that: In the specific strategies generated by the strategy generation unit, Prevention and control methods include spraying pesticides, pruning diseased branches, fertilizer control, burning and burying diseased plants, loosening the soil and draining water, and bagging young fruits; The implementation time for prevention and control measures refers to the optimal time to take preventive and control measures based on the assessed risk level. The duration of control measures refers to the estimated time required to suppress pests and diseases after taking control measures based on the assessed risk level. Prevention and control measures refer to the protective measures taken for normal plants to prevent the spread of the disease, depending on the specific prevention and control method used.

8. The real-time analysis and processing device for citrus big data based on deep learning according to claim 1, characterized in that: The visualization report module's 3D model report is based on preset plantation terrain information. It combines the 3D model reconstructed from the surface by feature extraction results from a convolutional neural network, pest and disease risk level assessment, and spread trend to obtain a visualized plantation scene model. The spread trend is then rendered as a time contour graphic overlaid on the model surface.

9. The real-time analysis and processing device for citrus big data based on deep learning according to claim 1, characterized in that: The bottom supplementary lighting device includes a reflective ring (2), the top wall of the reflective ring (2) is a reflective mirror (8), a rotating motor (3) is fixedly connected to the top wall of the reflective ring (2), the output shaft of the rotating motor (3) is vertically upward and coaxially fixedly connected to a transmission gear (4), the transmission gear (4) meshes with the outer circumference of a gear ring (5), a lamp ring (6) is coaxially fixedly connected to the top wall of the gear ring (5), the lamp ring (6) has an irregular shape along its axis, the outer circumference of the reflective ring (2) is fixedly connected to a chlorophyll excitation light source (7), the reflective ring (2) is fixedly sleeved with the middle part of the trunk (1) of the citrus plant, and both the gear ring (5) and the lamp ring (6) are slidably sleeved with the middle part of the trunk (1) of the citrus plant.

Images