Visual intelligent management system for strawberry planting

Abstract

The invention discloses a visualized intelligent management system for strawberry planting, and particularly relates to the technical field of agriculture, the system comprises a disease detection module, an environment monitoring module, a data analysis module and a remote management module; the disease detection module automatically identifies strawberry diseases through a deep learning technology; the environment monitoring module is used for collecting and monitoring environment parameters in the greenhouse in real time; the data analysis module is used for analyzing data provided by the disease detection and environment monitoring module; the remote management module supports remote monitoring and operation of greenhouse equipment. Through deep integration and intelligent design, the accuracy of disease detection, the precision of environment control and the overall efficiency of management are improved, and a more efficient, intelligent and user-friendly solution is provided for strawberry planting.

Description

Technical Field

[0001] The present invention relates to the field of agricultural technology, and more particularly to a visual intelligent management system for strawberry planting. Background Art

[0002] Strawberry is an important economic crop, and disease management is crucial during its cultivation. Common strawberry diseases include angular leaf spot, anthracnose, fruit rot, gray mold, leaf blight, leaf spot, and powdery mildew. If not detected and treated promptly, these diseases can seriously affect yield and fruit quality.

[0003] The existing open literature 1 (Strawberry disease recognition method based on improved MobileNet v3-Smal l model, 2024) proposed a strawberry disease recognition method based on the improved MobileNet v3-Smal l model, such as Figure 2 The figure shows the strawberry disease recognition process. This strawberry disease recognition method, based on an improved MobileNet v3-Smal l model, incorporates advanced technologies such as the Inception module, ULSAM, and CondConv. This enhances the model's ability to extract and focus on disease features, thereby improving recognition accuracy and generalization. However, this strawberry disease recognition method based on the improved MobileNet v3-Smal l model is prone to false detections in complex environments, making it difficult to meet the practical needs of strawberry cultivation.

[0004] The existing open document 2 (Design of Strawberry Disease Intelligent Recognition Terminal Based on Improved YOLOv5, 2024) proposes a strawberry disease intelligent recognition system based on the improved YOLOv5 deep learning model, such as Figure 3 As shown in the paper, by introducing the ECA attention mechanism to optimize the YOLOv5 model, an efficient and lightweight intelligent recognition terminal was designed. This strawberry disease intelligent recognition system, based on the improved YOLOv5 deep learning model, runs on an embedded device and can automatically capture and preprocess images in the field in real time, and identify the type and location of strawberry diseases. However, the performance of the embedded device affects real-time performance and accuracy, and lacks robustness in variable field environments.

[0005] Therefore, there is an urgent need for a visual intelligent management system for strawberry planting that can perform high-precision disease detection and real-time environmental monitoring, and comprehensively analyze disease and environmental data. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a visual intelligent management system for strawberry planting. By adopting a pre-trained ResNet152 model, a detection model based on a two-stage cascade is proposed, and high-precision strawberry planting disease detection is achieved through transfer learning. At the same time, temperature, humidity, light and soil moisture sensors are used to monitor environmental data in real time, grasp the greenhouse environmental status in real time, calculate the disease incidence and analyze the correlation between environmental parameters and diseases, so as to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A visual intelligent management system for strawberry planting includes a disease detection module, an environmental monitoring module, a data analysis module, and a remote management module. The disease detection module automatically identifies strawberry diseases using deep learning technology. The environmental monitoring module is used to collect and monitor environmental parameters in the greenhouse in real time. The data analysis module is used to analyze data provided by the disease detection and environmental monitoring modules. The remote management module supports remote monitoring and operation of greenhouse equipment.

[0009] The disease detection module adopts a two-stage cascade-based detection model to identify the type of strawberry disease. The two-stage cascade-based detection model includes a first stage and a second stage. The first stage adopts a confidence threshold of 0.3, first extracts the feature map of the input image, then generates a candidate bounding box through the region proposal network in the detection module, and then classifies the candidate bounding box. For areas classified as "abnormal", the first stage combines the non-maximum suppression technology to merge adjacent suspected lesion areas to generate new candidate areas, and inputs the new candidate areas into the second stage. The second stage adopts a confidence threshold of 0.7 and combines the feature pyramid network to extract multi-scale feature information. The second stage crops and adjusts the candidate areas of the first stage to a uniform size, and inputs them into the detection module of the second stage. The detection module of the second stage outputs the specific disease category and confidence of each candidate area, and maps the detection results back to the original image.

[0010] As a further solution of the present invention, the detection module includes a ResNet152 backbone network, a feature pyramid network, a region proposal network, a ROI pooling, a fully connected layer, a Softmax layer and object coordinates; the ResNet152 backbone network includes five convolution blocks, each of which outputs a feature map, the first convolution block outputs a feature map of size 112×112×64 to capture low-level features; the second convolution block outputs a feature map of size 56×56×256; the third convolution block outputs a feature map of size 28× 28×512; the fourth convolution block outputs a feature map of size 14×14×1024; the fifth convolution block outputs a feature map of size 7×7×2048; the feature pyramid network starts from the feature map of ResNet152 and adopts a top-down path: P5 comes directly from the fifth convolution block, with a feature map size of 7×7×2048; P4 combines the fourth convolution block and the features sampled from P5; P3 combines the third convolution block and the features sampled from P4; P2 combines the second convolution block and the features sampled from P3. The region proposal network slides a small network on the feature map, predicts the target score of the disease at each location, and generates bounding box coordinates for the predefined anchor box; the ROI pooling extracts a fixed-size feature map from the region proposal; the fully connected layer converts the pooled features into a format suitable for classification and regression; the softmax layer outputs the object category; the object coordinates output the bounding box coordinates of the disease region.

[0011] As a further embodiment of the present invention, an environmental monitoring module collects and monitors greenhouse environmental parameters in real time. Specifically, the module includes a temperature sensor, a humidity sensor, a light sensor, and a capacitive soil moisture sensor. The temperature sensor measures the greenhouse's temperature; the humidity sensor measures the greenhouse's humidity; the light sensor supports a wide range of light detection, capturing changes in greenhouse lighting conditions; and the capacitive soil moisture sensor monitors soil moisture. To ensure real-time and continuous data collection, the module sets the sensors to collect data every five minutes, which is then transmitted to a server via Wi-Fi.

[0012] As a further embodiment of the present invention, a data analysis module is used to analyze the data provided by the disease detection and environmental monitoring modules, including the following specific contents: the data analysis module converts the raw data collected by the disease detection module and the environmental monitoring module into actionable insights to help farmers optimize the health and production efficiency of strawberry crops. Specifically, it includes three steps: disease analysis, environmental analysis and recommendation generation.

[0013] The disease analysis aims to gain a deeper understanding of the occurrence patterns of strawberry diseases in the greenhouse. By processing the data provided by the disease detection module, the data analysis module can calculate the disease incidence rate, including the following steps:

[0014] Step S1: receiving image data containing disease information from a disease detection module.

[0015] Step S2: Identify strawberry plants in the image.

[0016] In step S3, the detected disease bounding boxes are matched with the corresponding plant bounding boxes. The degree of spatial overlap between the disease bounding boxes and the plant bounding boxes is calculated to determine which plant a particular disease belongs to. If the overlap between a disease bounding box and a plant bounding box exceeds a certain threshold, the plant is considered infected with the disease.

[0017] Step S4: For each disease type, count the number of plants infected with that disease. This is done by iterating through all plants and checking whether each plant has at least one bounding box associated with that disease type. If so, the plant is counted as infected with that disease.

[0018] Step S5, calculating the incidence of each disease based on the statistical results, using the following formula: disease incidence = number of plants infected with the disease / total number of plants.

[0019] Environmental analysis correlates the causal relationship between environmental parameters and disease occurrence. The data analysis module first uses sensors deployed within the greenhouse to collect real-time data on environmental parameters. Next, the module combines this data with historical disease occurrence records and uses regression analysis to quantify the relationship between variables like temperature and humidity and disease incidence, identifying patterns that significantly increase disease risk when certain environmental parameters reach specific ranges.

[0020] As a further solution of the present invention, the remote management module supports remote monitoring and operation of greenhouse equipment to help users achieve strawberry planting management, including the following specific contents: the remote management module includes hardware and software, allowing users to grasp the status of the greenhouse environment and adjust it anytime and anywhere. In terms of hardware, a series of controllable devices are deployed in the greenhouse, such as ventilation fans, sprinkler systems, heaters and sunshade nets, etc. The devices can be turned on or off or adjusted according to remote instructions to cope with environmental changes in the greenhouse or user needs. In terms of software, the remote management module has developed mobile and PC applications, and users can access the system through smartphones, tablets or computers. The application is connected to the cloud server, and users can view the environmental parameters in the greenhouse in real time, such as observing the changing trends of data such as temperature and humidity through charts and dashboards. In addition, the remote management module has a disease early warning function. When an environmental anomaly or disease risk is detected, a notification will be pushed immediately through the application, and the user can take timely measures accordingly. The user can also remotely turn on and off and adjust the equipment through the application.

[0021] The technical effects and advantages of the present invention's visual intelligent management system for strawberry planting: The present invention adopts a pre-trained ResNet152 model (PlantNet), accumulates rich plant field knowledge on the PlantCLEF dataset, and realizes high-precision disease detection through transfer learning. Using temperature, humidity, light and soil moisture sensors, data is collected every 5 minutes to grasp the greenhouse environmental status in real time, optimize growth conditions and reduce resource waste. The data analysis module converts disease and environmental data into actionable insights, calculates the incidence of diseases and analyzes the correlation between environmental parameters and diseases, helping farmers make scientific decisions. The remote management module realizes real-time monitoring and operation of the greenhouse through a Wi-Fi network. Compared with the shortcomings of the existing technology in a changeable field environment, the present invention improves the accuracy of disease detection, the precision of environmental control and the overall efficiency of management through deep integration and intelligent design, providing a more efficient, intelligent and user-friendly solution for strawberry planting, thereby improving yield and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a structural diagram of a visual intelligent management system for strawberry planting according to the present invention.

[0023] Figure 2 The present invention is a flowchart of strawberry pest and disease identification in the prior art.

[0024] Figure 3 This is the structural diagram of the strawberry disease intelligent identification system based on the improved YOLOv5 deep learning model in the existing technology.

[0025] Figure 4 Schematic diagram of the structure of the detection module of the present invention.

[0026] Figure 5 This is an example of a noise-free strawberry sample in the PlantCLEF dataset of the present invention.

[0027] Figure 6 This is a schematic diagram of the structure of the detection model based on two-stage cascade proposed in the present invention.

[0028] Figure 7 These are areas classified as “abnormal” in the first stage of the present invention.

[0029] Figure 8 This is the merged area that is input to the second stage of the present invention.

[0030] Figure 9 This is a sample diagram of the final detection results. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] Example 1

[0033] See Figure 1 The structural schematic diagram shown in the figure shows a visual intelligent management system for strawberry planting of the present invention, which includes a disease detection module, an environmental monitoring module, a data analysis module and a remote management module; the disease detection module automatically identifies strawberry diseases through deep learning technology; the environmental monitoring module is used to collect and monitor environmental parameters in the greenhouse in real time; the data analysis module is used to analyze data provided by the disease detection and environmental monitoring modules; and the remote management module supports remote monitoring and operation of greenhouse equipment.

[0034] Furthermore, the disease detection module automatically identifies strawberry diseases through deep learning technology, including: the disease detection module uses the ResNet152 model pre-trained on the PlantCLEF 2017 dataset, named PlantNet, as the backbone network for feature extraction of the detection module, such as Figure 4As shown, the detection module includes a ResNet152 backbone network, a Feature Pyramid Network (FPN), a Region Proposal Network (RPN), ROI pooling, a fully connected layer, a Softmax layer, and object coordinates. The ResNet152 backbone extracts features from the input image, while the Feature Pyramid Network generates multi-scale feature representations for detecting disease regions of varying sizes. These features are then fed into the Region Proposal Network (RPN) and subsequent layers, ultimately outputting the disease category and location. The ResNet152 backbone consists of five convolutional blocks, each of which outputs a feature map. The first convolutional block outputs a feature map of size 112×112×64, capturing low-level features; the second convolutional block outputs a feature map of size 56×56×256; the third convolutional block outputs a feature map of size 28×28×512; the fourth convolutional block outputs a feature map of size 14×14×1024; and the fifth convolutional block outputs a feature map of size 7×7×2048, capturing high-level features. The feature pyramid network starts from the feature map of ResNet152 and adopts a top-down path: P5 comes directly from the fifth convolutional block, with a feature map size of 7×7×2048; P4 combines the fourth convolutional block and features sampled from P5; P3 combines the third convolutional block and features sampled from P4; P2 combines the second convolutional block and features sampled from P3. The region proposal network slides a small network on the feature map, predicts the target score of the disease at each location, and generates bounding box coordinates for predefined anchor boxes. The ROI pooling extracts a fixed-size feature map from the region proposal. The fully connected layer converts the pooled features into a format suitable for classification and regression. The softmax layer outputs the object category. The object coordinates output the bounding box coordinates of the disease region.

[0035] like Figure 5 As shown in the figure, the PlantCLEF 2017 dataset contains approximately 1.25 million high-quality plant images, covering leaves, fruits, flowers and other parts of a variety of plants. The data is rich in content and highly relevant to the plant field. In addition, the ResNet152 model, as a deep residual network, effectively alleviates the gradient vanishing problem by introducing residual connections, and can maintain stable performance even when processing complex images. The PlantNet can overcome the problem of insufficient labeled data: in the task of strawberry disease detection, obtaining large-scale labeled datasets is costly and time-consuming, while PlantNet captures rich plant domain knowledge in advance by pre-training on the PlantCLEF dataset. This domain knowledge transfer learning method enables the system to achieve efficient feature extraction and model fine-tuning even with only a small amount of strawberry disease labeled data, thereby significantly improving detection performance.

[0036] like Figure 6As shown, the disease detection module proposes a two-stage cascade detection model, inspired by the behavior of human experts in disease identification: first, a rapid scan of the plant is performed to identify suspicious areas, followed by a detailed inspection of these areas to confirm the specific disease type. This cascade strategy not only improves detection accuracy but also reduces computing resource consumption through phased processing.

[0037] The goal of the first stage of the two-stage cascade-based detection model is to quickly identify suspicious disease areas in strawberry plants and preliminarily classify the suspicious disease areas into "normal", "abnormal (suspected lesions)" and "background objects". In order to ensure that potential disease areas are not missed, a lower confidence threshold (0.3) is used in the first stage to improve the recall rate. The first stage of the two-stage cascade-based detection model first uses PlantNet to extract the feature map of the input image, and then generates a series of candidate bounding boxes through the region proposal network (RPN) in the detection module. The region proposal network slides the window on the feature map, predicts whether the anchor point at each position contains an object, and performs preliminary regression adjustment on the bounding box. Subsequently, the candidate bounding box is classified to determine whether it belongs to "normal", "abnormal" or "background object" (such as support columns, plastic films, etc. in greenhouses). Figure 7 As shown in , for regions classified as “abnormal”, the first stage combines the non-maximum suppression (NMS) technique, as Figure 8 As shown, adjacent suspected lesion regions are merged to generate new larger candidate regions, while background objects are explicitly excluded, and the merged suspected lesion regions are input into the second stage.

[0038] The second-stage detection network of the two-stage cascade-based detection model performs fine classification on the candidate areas generated in the first stage and identifies specific disease types, including angular leaf spot, anthracnose fruit rot, gray mold, leaf blight, leaf spot and powdery mildew (fruit / leaves). The second stage adopts a higher confidence threshold (0.7) and combines the feature pyramid network (FPN) to extract multi-scale feature information. The feature pyramid network can effectively process disease features of different scales by constructing a top-down feature pyramid structure, especially when the size of disease symptoms varies greatly (such as small spots on leaves and large areas of rot on fruits), thereby improving detection performance. The second stage crops and adjusts the candidate areas of the first stage to a uniform size, and then inputs them into the detection module of the second stage. The second-stage detection module is also based on the PlantNet backbone network, generates multi-scale feature maps through the feature pyramid network, and then uses the region proposal network and classifier to finely locate and classify the candidate areas. Finally, as Figure 9As shown, the detection module in the second stage outputs the specific disease category and confidence level of each candidate area, and maps the detection results back to the original image to be intuitively presented to the user.

[0039] In this embodiment, the present invention collects 4,560 images from different greenhouses, and the images cover the conditions of strawberries under various environmental conditions, including the early, middle and late stages of the disease, as well as healthy strawberry plants. The image data set is divided into a training set, a validation set and a test set in a ratio of 7:2:1, and data enhancement is performed by geometric transformations (such as resizing, cropping, rotation, flipping and color adjustment), wherein the training and validation images are enhanced to 39,708, and the test images retain the original 1,416. The initial learning rate of the detection model based on the two-stage cascade is 0.001, the maximum number of iterations is 250,000 times, and at the 80,000th iteration, the learning rate is multiplied by 0.1 for attenuation, and at the 150,000th iteration, the learning rate is multiplied by 0.1 for attenuation; the momentum rate is 0.9. Table 1 shows the detection performance comparison results of the detection model based on the two-stage cascade and the single-stage detector (using ResNet152 as the backbone network and combining the feature pyramid network structure). Table 2 shows the average precision (AP) of the cascade detector pre-trained with PlantNet for different disease categories of strawberry.

[0040] Table 1 Detection results of single-stage detector and cascade detector

[0041] Detector type mAP Cascade detector 91.65％ Single-stage detector 86.40％

[0042] Table 2 Average precision of different disease categories

[0043] Disease Type Average Precision (AP) Angular spot disease 94.56％ Anthracnose fruit rot 95.01％ Gray mold 94.52％ Leaf blight 91.82％ Leaf spot 87.83％ powdery mildew 95.36％

[0044] Table 1 shows the mean average precision (mAP) of single-stage detectors and cascade detectors on the PlantNet pre-training dataset.

[0045] Furthermore, the environmental monitoring module collects and monitors environmental parameters within the greenhouse in real time. The module includes a temperature sensor, a humidity sensor, a light sensor, and a capacitive soil moisture sensor. The temperature sensor measures the temperature within the greenhouse; the humidity sensor measures the humidity within the greenhouse; the light sensor supports a wide range of light detection, capturing changes in lighting conditions within the greenhouse; and the capacitive soil moisture sensor monitors soil moisture. To ensure real-time and continuous data, the environmental monitoring module sets the sensor to collect data every five minutes. This frequency allows for timely reflection of environmental changes while avoiding the waste of resources associated with overly frequent data collection. The collected data is transmitted to the server via Wi-Fi.

[0046] Furthermore, the data analysis module is used to analyze the data provided by the disease detection and environmental monitoring modules, including the following specific contents: the data analysis module converts the raw data collected by the disease detection module and the environmental monitoring module into actionable insights to help farmers optimize the health and production efficiency of strawberry crops, specifically including three steps: disease analysis, environmental analysis and recommendation generation.

[0047] The disease analysis aims to gain a deeper understanding of the occurrence patterns of strawberry diseases in the greenhouse. By processing the data provided by the disease detection module, the data analysis module can calculate the disease incidence rate, including the following steps:

[0048] Step S1: receiving image data containing disease information from a disease detection module, such as a set of strawberry field photos with disease boundary boxes marked.

[0049] Step S2: Identify strawberry plants in the image.

[0050] In step S3, the detected disease bounding boxes are matched with the corresponding plant bounding boxes. The degree of spatial overlap (e.g., Intersection over Union (IoU)) between the disease bounding boxes and the plant bounding boxes is calculated to determine which plant a disease belongs to. If the overlap between a disease bounding box and a plant bounding box exceeds a certain threshold (e.g., IoU > 0.5), the plant is considered infected with the disease.

[0051] Step S4: For each disease type, the number of plants infected with the disease is counted. The specific method is to traverse all plants and check whether each plant has at least one bounding box belonging to the disease type associated with it. If so, the plant is counted as infected with the disease.

[0052] Step S5, calculating the incidence of each disease based on the statistical results, using the following formula: disease incidence = number of plants infected with the disease / total number of plants.

[0053] In step S6, the calculated disease incidence is presented to farmers in an intuitive form, for example, through a numerical table (angular leaf spot: 20%, anthracnose fruit rot: 15%, powdery mildew: 10%) or a visual chart (such as a bar chart), to help them quickly understand the distribution and severity of the disease.

[0054] Environmental analysis links causal relationships between environmental parameters and disease occurrence. The data analysis module first uses sensors deployed within the greenhouse to collect real-time data on these parameters. Next, the module combines this collected environmental parameter data with historical disease occurrence records and uses regression analysis to quantify the relationship between variables such as temperature and humidity and disease incidence, identifying patterns in which disease risk increases significantly when certain environmental parameters reach specific ranges. This analysis allows the module to identify key conditions. For example, the probability of gray mold occurrence increases significantly when humidity exceeds 80% and temperatures are maintained between 15-25°C. This is because high humidity and moderate temperatures provide an ideal breeding environment for fungal spores. Similarly, excessive soil moisture can lead to root diseases, while insufficient light can weaken plant resistance, indirectly promoting disease development. By establishing causal relationships between these environmental parameters and diseases, the module not only identifies specific risk thresholds but also monitors whether current or future environmental conditions are approaching these thresholds. Once potential risks are detected, the system issues timely warnings, enabling growers to take preventative measures (such as increasing ventilation, reducing humidity, or adjusting irrigation) to effectively prevent disease occurrence.

[0055] Furthermore, the remote management module supports remote monitoring and operation of greenhouse equipment, helping users to manage strawberry cultivation, including: the remote management module includes hardware and software, allowing users to monitor the greenhouse environment and adjust it anytime and anywhere. In terms of hardware, a series of controllable devices are deployed in the greenhouse, such as ventilation fans, sprinkler systems, heaters and sunshade nets. These devices can be turned on or off or adjusted according to remote commands to respond to changes in the greenhouse environment or user needs. For example, in high temperature conditions, users can remotely turn on ventilation fans to cool down, or start sprinkler systems for irrigation when the soil is dry. At the same time, the greenhouse is also equipped with a sensor network, including temperature, humidity, light intensity and soil moisture sensors, which collect environmental data in real time and transmit it to the cloud server via wireless network, providing a reliable basis for remote monitoring. In terms of software, the remote management module has developed mobile and PC applications, and users can access the system through smartphones, tablets or computers. The applications are connected to the cloud server, and users can view the environmental parameters in the greenhouse in real time, such as observing the changing trends of data such as temperature and humidity through charts and dashboards. The remote management module also features a disease warning function. When environmental anomalies or disease risks are detected, notifications are sent immediately via the app, allowing users to take timely action, such as adjusting ventilation or irrigation settings. Users can also remotely turn devices on and off and adjust them through the app, such as activating a dehumidifier when humidity is too high or adjusting shade screens when light levels are low.

[0056] The present invention adopts the pre-trained ResNet152 model (PlantNet), accumulates rich plant domain knowledge on the PlantCLEF dataset, and achieves high-precision disease detection through transfer learning. Using temperature, humidity, light and soil moisture sensors, data is collected every 5 minutes to grasp the greenhouse environment status in real time, optimize growth conditions and reduce resource waste. The data analysis module converts disease and environmental data into actionable insights, calculates the incidence of diseases and analyzes the correlation between environmental parameters and diseases, helping farmers make scientific decisions. The remote management module realizes real-time monitoring and operation of the greenhouse through the Wi-Fi network. Compared with the shortcomings of the existing technology in the changeable field environment, the present invention improves the accuracy of disease detection, the precision of environmental control and the overall efficiency of management through deep integration and intelligent design, providing a more efficient, intelligent and user-friendly solution for strawberry cultivation, thereby improving yield and quality.

[0057] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0058] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A visual intelligent management system for strawberry planting, characterized in that: It includes a disease detection module, an environmental monitoring module, a data analysis module, and a remote management module; the disease detection module automatically identifies strawberry diseases using deep learning technology; the environmental monitoring module is used to collect and monitor environmental parameters in the greenhouse in real time; the data analysis module is used to analyze data provided by the disease detection and environmental monitoring modules; and the remote management module supports remote monitoring and operation of greenhouse equipment. The disease detection module uses a two-stage cascade-based detection model to identify strawberry disease types. The two-stage cascade-based detection model includes a first stage and a second stage. The first stage uses a confidence threshold of 0.3, first extracts a feature map of the input image, then generates candidate bounding boxes through a region proposal network in the detection module, and then classifies the candidate bounding boxes. For areas classified as "abnormal", the first stage combines non-maximum suppression technology to merge adjacent suspected lesion areas to generate new candidate regions, which are input into the second stage. The second stage uses a confidence threshold of 0.7 and combines it with a feature pyramid network to extract multi-scale feature information. The second stage crops and adjusts the candidate areas in the first stage to a uniform scale and inputs them into the detection module in the second stage. The detection module in the second stage outputs the specific disease category and confidence level of each candidate area and maps the detection results back to the original image.

2. A strawberry planting visualization intelligent management system according to claim 1, characterized in that The detection module includes a ResNet152 backbone network, a feature pyramid network, a region proposal network, a ROI pooling layer, a fully connected layer, a Softmax layer, and object coordinates; the ResNet152 backbone network includes five convolution blocks, each of which outputs a feature map. The first convolution block outputs a feature map of size 112×112×64 to capture low-level features; the second convolution block outputs a feature map of size 56×56×256; the third convolution block outputs a feature map of size 28×28×51 2; the fourth convolution block outputs a feature map of size 14×14×1024; the fifth convolution block outputs a feature map of size 7×7×2048; the feature pyramid network starts from the feature map of ResNet152 and adopts a top-down path: P5 comes directly from the fifth convolution block, with a feature map size of 7×7×2048; P4 combines the fourth convolution block and the features sampled from P5; P3 combines the third convolution block and the features sampled from P4; P2 combines the second convolution block and the features sampled from P3.

3. A strawberry planting visualization intelligent management system according to claim 2, characterized in that The region proposal network slides a small network on the feature map to predict the target score of the disease at each location and generate bounding box coordinates for the predefined anchor box; the ROI pooling extracts a fixed-size feature map from the region proposal; the fully connected layer converts the pooled features into a format suitable for classification and regression; the Softmax layer outputs the object category; the object coordinates output the bounding box coordinates of the disease area.

4. A visual intelligent management system for strawberry planting according to claim 1, characterized in that, The dataset of the detection module contains 4,560 images, which are divided into training set, validation set and test set in a ratio of 7:2:

1. After geometric transformation enhancement, the number of training / validation images reaches 39,708. The initial learning rate of the model is 0.001, the maximum iteration is 250,000 times, and the momentum rate is 0.

9.

5. A visual intelligent management system for strawberry planting according to claim 1, characterized in that, The environmental monitoring module includes a temperature sensor, a humidity sensor, a light sensor, and a capacitive soil moisture sensor. It collects data every 5 minutes and transmits it to the server via Wi-Fi to monitor the temperature, humidity, light intensity, and soil moisture in the greenhouse in real time.

6. A visual intelligent management system for strawberry planting according to claim 1, characterized in that, The disease analysis steps of the data analysis module include: Step S1, receiving image data containing disease information from a disease detection module; Step S2, identifying strawberry plants in the image; Step S3, matching the detected disease bounding box with the corresponding plant bounding box; Step S4, for each disease type, counting the number of plants infected with the disease; Step S5: Calculate the incidence of each disease based on the statistical results.

7. A visual intelligent management system for strawberry planting according to claim 1, characterized in that, The environmental analysis of the data analysis module uses regression analysis to correlate temperature, humidity and soil moisture parameters with disease incidence, identify key risk conditions, set thresholds and trigger early warnings.

8. A visual intelligent management system for strawberry planting according to claim 1, characterized in that, The remote management module includes hardware and software. The hardware includes ventilation fans, sprinkler systems, heaters, and sunshade net equipment; the software includes mobile and PC applications that support real-time viewing of environmental data, receiving disease warnings, and remote adjustment of equipment parameters.

9. A visual intelligent management system for strawberry planting according to claim 1, characterized in that, The environmental analysis establishes a causal relationship between environmental parameters and diseases. For example, high soil moisture can easily cause root diseases, and insufficient light weakens plant resistance. Based on this, ventilation and irrigation adjustment recommendations are generated.

Images