Aloe disease grading system and method based on improved Deeplab v3 +

Abstract

The invention discloses an aloe disease grading system and method based on improved Deeplab v3 +. The system comprises a data acquisition module, a data preprocessing module, a semantic segmentation module, a performance estimation module, an input module, an identification module, a grading module, an output module and a database. The semantic segmentation module is used for carrying out feature extraction on aloe leaves and surface scabs to obtain aloe leaf recognition and scabs segmentation results; the semantic segmentation module adopts a UAMT network framework and comprises a teacher model and a student model, and constraints on uncertain information are added in a'prediction consistency 'part, so that when the consistency loss is calculated, only a result with relatively small prediction deviation of the teacher model and the student model is learned, and a result with relatively large prediction deviation is inhibited; both the teacher model and the student model adopt an improved Deeplab v3 + network; according to the method, segmentation and identification of the aloe leaf surface diseases and grading of the disease degrees can be realized, and the method has relatively high precision and efficiency.

Description

Technical Field

[0001] The invention belongs to the technical field of aloe disease segmentation, and particularly relates to an aloe disease grading system and method based on improved Deeplab v3+. Background Art

[0002] Aloe vera is a succulent plant widely distributed around the world. It is not only edible, but also has many application values ​​in medicine and beauty. At the same time, the global market size of aloe vera continues to expand. However, aloe vera is also very susceptible to diseases. Aloe vera diseases, mainly leaf spot and anthracnose, seriously affect the yield of aloe vera. In modern agricultural production, the timely identification and accurate classification of crop diseases are crucial to crop health and yield. Especially in aloe vera cultivation, the management of diseases directly affects its economic benefits and quality stability. If the disease is not discovered in time and the degree of disease is not evaluated and effectively controlled, it will cause losses. Therefore, it is of great significance to segment aloe vera diseases and classify the degree of disease.

[0003] At present, traditional disease identification methods mainly rely on manual observation and empirical judgment, which is highly subjective, inefficient and has low diagnostic accuracy. In addition, for large-scale aloe field disease surveys, traditional methods are not only very time-consuming and labor-intensive, but also prone to errors, making it difficult to achieve the desired effect. Summary of the invention

[0004] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and to provide an aloe disease grading system and method based on improved Deeplabv3+, which uses the improved Deeplab v3+ semantic segmentation algorithm to segment aloe leaves and their surface lesions, thereby realizing the segmentation and identification of aloe leaf surface diseases and the grading of their disease severity, with high accuracy and efficiency, and can realize the identification and grading research of aloe diseases, effectively solving the shortcomings of manual identification.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] The present invention provides an aloe disease grading system based on improved Deeplab v3+, comprising a data acquisition module, a data preprocessing module, a semantic segmentation module, a performance estimation module, an input module, a recognition module, a grading module, an output module and a database;

[0007] The data acquisition module is used to obtain the leaf data of aloe vera taken from above or from the side;

[0008] The data preprocessing module preprocesses the collected aloe data video to obtain a picture data set, and marks the aloe leaves and the disease spots on the leaf surfaces, and finally divides the data set;

[0009] The semantic segmentation module is used to extract features of aloe leaves and surface lesions to obtain aloe leaf recognition and lesion segmentation results; the semantic segmentation module adopts the UAMT network framework, which includes a teacher model Tnet and a student model Snet, and adds constraints on uncertain information in the "prediction consistency" part, so that when calculating the consistency loss, only the results with smaller prediction deviations of the teacher model Tnet and the student model Snet are learned, while the results with larger prediction deviations are suppressed; the teacher model Tnet and the student model Snet have the same structure, and both use the improved Deeplab v3+ network; the UAMT network framework uses the improved Deeplab v3+ network for deep learning in the process of aloe plant detection, and improves the confidence of the network output by evaluating the uncertainty in the prediction process of the teacher network; the lightweight and efficient MobileNetV2 architecture is used to optimize the model backbone network, and a multi-scale attention module is introduced into the backbone network and the CBAM attention mechanism is inserted to improve the efficiency and real-time performance of the semantic segmentation module;

[0010] The performance estimation module is used to evaluate the detection results, test the MIoU, MPA and accuracy of the detection results, and return to the semantic segmentation module if it is lower than the set threshold, improve the solution, and re-extract features;

[0011] The input module is used for inputting the picture information required for aloe disease detection into a model;

[0012] The recognition module is used to recognize and segment the aloe leaves and the diseased spots on their surfaces in the input video or image;

[0013] The grading module is used to calculate the segmentation results of the recognition module and grade the degree of aloe disease;

[0014] The output module is used to output and present the disease identification and classification data to the terminal display device, and at the same time output the calculated results in a weighted manner;

[0015] The database is used to store the characteristics of aloe leaves and surface disease spots.

[0016] As a preferred technical solution, the method of obtaining the aloe leaf data shot from above or from the side is as follows:

[0017] Use a drone or handheld terminal to take photos to obtain video data of aloe leaves from overhead or side shots.

[0018] As a preferred technical solution, the preprocessing is mean normalization, and the feature scaling formula is:

[0019]

[0020] In the formula, x′ is the average normalized value, x is the original value; max(x) represents the maximum value of the sample data, and min(x) represents the minimum value of the sample data.

[0021] As a preferred technical solution, the improved Deeplab v3+ network includes an encoder and a decoder, the multi-scale attention module MSA introduces a spatial channel attention module CBAM on features of different scales, and then fuses enhanced feature representations of different scales to further enhance the representation capabilities of feature maps of different scales; the Deeplab v3+ network selects the 4th and 8th layers in the backbone network MobilenetV2 as key layers for feature extraction, which correspond to 1 / 4 and 1 / 16 scales of the input image, respectively, and have initial channel numbers of 24 and 64, respectively, effectively capturing different levels of features from fine texture to high-level semantics; to ensure the consistency of feature maps in the spatial dimension in subsequent processing, the smaller 8th layer feature map is first upsampled by 4 times to match the spatial resolution of the 4th layer feature map, and then the sampled feature maps are respectively input into the CBAM module; finally, the two feature maps enhanced by the CBAM module are effectively combined together, and the combined feature map contains feature information of different levels and has richer semantic and spatial information.

[0022] As a preferred technical solution, the CBAM module can adaptively adjust the feature map in the convolutional neural network. When an intermediate feature map is given, the CBAM module calculates the attention map along the two dimensions of channel and space, and then applies these attention maps to the input feature map for adaptive feature improvement.

[0023] As a preferred technical solution, the performance estimation module compares the received data with the feature extraction results of the target detection module, and the results include four parts: TP is the positive class determined as the positive class, TN is the negative class determined as the negative class, FN is the positive class determined as the negative class, and FP is the negative class determined as the positive class;

[0024] MIoU, MPA and accuracy were selected as evaluation indicators of the algorithm;

[0025] The calculation formula of MIoU is:

[0026] The calculation formula of MPA is:

[0027] The accuracy is calculated as follows:

[0028] The closer all evaluation indicators are to 1, the better the effect.

[0029] In a second aspect, the present invention provides an aloe disease grading method based on improved Deeplab v3+, comprising the following steps:

[0030] S1, obtaining the leaf data of aloe vera taken from top or side;

[0031] S2, pre-process the captured video data, capture a video screenshot every 0.5-1s through the Python program, filter out the available pictures and randomly distribute them into training sets and test sets according to the set ratio;

[0032] S3. Use the Labelme tool to label the aloe objects in the experimental data, convert the output json file into a png image in VOC format for storage. The different color classifications of the pixels in the image represent the healthy leaves, diseased spots and filtered background of the aloe.

[0033] S4, put the annotated data into the semantic segmentation module, perform deep learning on the improved Deeplab v3+, use MIoU, MPA and accuracy as evaluation indicators of the algorithm, if the evaluation indicator is less than the set threshold, then improve the scheme, retrain the detection model with the wrong recognition data, and store the corrected deep learning results in the model library, so as to achieve the segmentation detection effect of aloe leaf diseases;

[0034] S5. The user inputs the required aloe leaf images through the visualization platform. The model can segment the lesions on the surface of the aloe leaves and realize the classification study of aloe diseases based on the calculation of the proportion of the lesion area to the total leaf area.

[0035] As a preferred technical solution, step S1 is specifically:

[0036] Operate the drone to fly above the aloe vera field with the flight altitude set to 1.5 to 2.5 meters and the camera deflection angle set to be perpendicular to the ground to take a bird's-eye view of the aloe vera field, and record video data at a horizontal flight speed of less than 1 meter per second; or manually shuttle through the aloe vera field and use a mobile phone to take direct shots of the surface of the aloe leaves, with shooting angles including vertical downward shooting from above and 45 degrees diagonal shooting from the side.

[0037] As a preferred technical solution, in step S4, deep learning is performed on the improved Deeplab v3+, specifically:

[0038] S41, putting the labeled data into a semantic segmentation module, where the semantic segmentation module is obtained based on an improved Deeplab v3+ model;

[0039] S42 and DeepLab use the convolutional neural network (CNN) architecture, which is based on the improvement of the selected standard network and introduces dilated convolution, which enables the network to extract features at different scales without losing resolution.

[0040] S43. During the training process, DeepLab uses the cross entropy loss function to evaluate the difference between the prediction and the true label, and the model will continuously adjust the parameters based on the feedback of the loss.

[0041] S44. After training, the test set is used to evaluate the performance of the model. Indicators such as intersection over union (IoU) and pixel accuracy are usually used to measure the segmentation quality of the model.

[0042] As a preferred technical solution, in step S4, the threshold is 0.9.

[0043] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0044] The present invention provides an aloe disease degree grading method based on deep learning. The UAMT semi-supervised learning architecture is adopted, and the deep network learning architecture of Deeplab v3+ in the prior art is used to perform deep learning in the process of aloe plant detection. The uncertainty in the prediction process of the teacher network is evaluated, and the confidence of the network output is improved. The lightweight and efficient MobileNetV2 architecture is used to optimize the model backbone network, and a multi-scale attention module is introduced into the backbone and a CBAM attention mechanism is inserted, which improves the efficiency and real-time performance of the model, has a strong learning effect, and the accuracy and efficiency are higher than those of general manual observation and subjective judgment.

[0045] The present invention combines the automatic image collection of unmanned aerial vehicles to segment the aloe leaves and the lesions on their surfaces, and calculates the areas of the two and their proportions, thereby analyzing the aloe disease results. It has high pixel accuracy and F1 value, and can realize disease monitoring and disease severity rating of the aloe growth condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 It is a structural diagram of the aloe disease grading system based on the improved Deeplab v3+ of the present invention;

[0048] Figure 2 It is the network architecture diagram of UAMT of the present invention;

[0049] Figure 3 It is a module structure diagram of the CBAM attention mechanism of the present invention;

[0050] Figure 4 This is the flowchart of the multi-scale attention module and Deeplabv3+ network of the present invention. DETAILED DESCRIPTION

[0051] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0052] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments.

[0053] like Figure 1 As shown, this embodiment provides an aloe disease grading system based on improved Deeplab v3+, including a data acquisition module, a data preprocessing module, a semantic segmentation module, a performance estimation module, an input module, a recognition module, a grading module, an output module and a database;

[0054] The data acquisition module is used to obtain the leaf data of aloe vera taken from above or from the side;

[0055] The data preprocessing module preprocesses the collected aloe data video to obtain a picture data set, and marks the aloe leaves and the disease spots on the leaf surfaces, and finally divides the data set;

[0056] The semantic segmentation module is used to extract features of aloe leaves and surface lesions to obtain aloe leaf recognition and lesion segmentation results; the semantic segmentation module adopts the UAMT network framework, the UAMT framework includes a teacher model Tnet and a student model Snet, and adds constraints on uncertain information in the "prediction consistency" part, so that when calculating the consistency loss, only the results with smaller prediction deviations of the teacher model Tnet and the student model Snet are learned, while the results with larger prediction deviations are suppressed; the teacher model Tnet and the student model Snet have the same structure, and both use the improved Deeplab v3+ network.

[0057] The performance estimation module is used to evaluate the detection results, test the MIoU, MPA and accuracy of the detection results, and return to the semantic segmentation module if it is lower than the set threshold, improve the solution, and re-extract features;

[0058] The input module is used for inputting the picture information required for aloe disease detection into a model;

[0059] The recognition module is used to recognize and segment the aloe leaves and the diseased spots on their surfaces in the input video or image;

[0060] The grading module is used to calculate the segmentation results of the recognition module and grade the degree of aloe disease;

[0061] The output module is used to output and present the disease identification and classification data to the terminal display device, and at the same time output the calculated results in a weighted manner;

[0062] The database is used to store the characteristics of aloe leaves and surface disease spots.

[0063] Furthermore, in the data acquisition module, the step of obtaining the aloe leaf data shot from above or from the side is as follows:

[0064] Image data of aloe leaves can be collected by using drones for aerial photography or by using handheld devices such as smartphones and tablets to photograph aloe plants on the ground. These images can be static photos or dynamic video materials, aiming to comprehensively record the morphological characteristics of aloe leaves at different angles for subsequent analysis of their growth status, health status and other information. In this way, large areas of aloe vera planting can be efficiently monitored, and data can be obtained regularly to track the growth and changes of aloe vera.

[0065] In a more specific embodiment, a drone is operated to fly above an aloe vera field with a flight altitude set to 1.5 meters to 2.5 meters and a camera deflection angle set to be perpendicular to the ground to take a bird's-eye view of the aloe vera field, and to shoot and record video data at a horizontal flight speed of less than 1 meter per second; or people manually shuttle through the aloe vera field and use a mobile phone to take direct shots of the surface of the aloe leaves, with shooting angles including vertical downward shooting from above and 45 degrees diagonal downward shooting from the side.

[0066] Furthermore, in the data preprocessing module, the preprocessing is mean normalization, and the feature scaling formula is:

[0067]

[0068] In the formula, x′ is the average normalized value, x is the original value; max(x) represents the maximum value of the sample data, and min(x) represents the minimum value of the sample data.

[0069] Furthermore, in the semantic segmentation module, the UAMT network framework adopts the improved Deeplab v3+ network to perform deep learning in the process of aloe plant detection, and improves the confidence of the network output by evaluating the uncertainty in the teacher network prediction process; the lightweight and efficient MobileNetV2 architecture is used to optimize the model backbone network, and a multi-scale attention module is introduced into the backbone network and the CBAM attention mechanism is inserted to improve the efficiency and real-time performance of the semantic segmentation module.

[0070] Furthermore, Figure 2 As shown in the figure, the Uncertainty-Aware Mean Teacher (UAMT) network architecture adds constraints on uncertain information in the "prediction consistency" part while maintaining the core idea of ​​the Mean Teacher framework. Adding constraints on this part allows only the results with smaller prediction deviations between the two networks to be learned when calculating the consistency loss, while suppressing the results with larger prediction deviations. Therefore, Snet is optimized and supervised more reliably, which in turn encourages the Tnet part to make higher quality predictions.

[0071] The teacher model Tnet and the student model Snet have the same structure, and both are based on the improved DeeplabV3+ network. After the labeled data XL is input into Snet, ysL is output, and supervised learning is performed with the label Y. The loss function is the cross entropy loss function, denoted as Lsup. After the unlabeled data XU is input into Snet and Tnet, ysU and ytU are predicted respectively, and the consistency loss between the two is denoted as Lcon.

[0072] like Figure 3As shown in Figure 2, the CBAM module is able to adaptively adjust the feature maps in the convolutional neural network. Given an intermediate feature map, the CBAM module calculates attention maps along the channel and spatial dimensions, and then applies these attention maps to the input feature map for adaptive feature improvement.

[0073] like Figure 4 As shown in the figure, the multi-scale attention module (MSA) introduces a spatial channel attention module (CBAM) on features of different scales, and then fuses enhanced feature representations of different scales to further enhance the representation capabilities of feature maps of different scales. Specifically, it selects the 4th and 8th layers in the backbone network MobilenetV2 (17 layers) as the key layers for feature extraction. These two layers correspond to the 1 / 4 and 1 / 16 scales of the input image, respectively, and their initial number of channels is 24 and 64, respectively, effectively capturing different levels of features from fine texture to high-level semantics. To ensure the consistency of feature maps in the spatial dimension in subsequent processing, the smaller 8th layer feature map (1 / 16 scale) is first upsampled 4 times to match the spatial resolution of the 4th layer feature map (1 / 4 scale), and then input into the CBAM module respectively. Finally, the two feature maps enhanced by the CBAM module are effectively combined, which contain feature information at different levels and have richer semantic and spatial information.

[0074] Furthermore, in the performance estimation module, the received data is compared with the feature extraction results of the target detection module, and the results include four parts: TP is the positive class judged as positive, TN is the negative class judged as negative, FN is the positive class judged as negative, and FP is the negative class judged as positive;

[0075] MIoU, MPA and accuracy were selected as evaluation indicators of the algorithm;

[0076] The calculation formula of MIoU is:

[0077] The calculation formula of MPA is:

[0078] The accuracy is calculated as follows:

[0079] The closer all evaluation indicators are to 1, the better the effect.

[0080] Furthermore, improvement methods include but are not limited to: changing the ratio of teacher model and student model labels in the Mean teacher model; changing the attention mechanism; using other loss functions, etc. This article only provides a more preferred solution in aloe lesion segmentation.

[0081] In another embodiment of the present application, a method for grading aloe diseases based on improved Deeplab v3+ is provided, the method comprising the following steps:

[0082] S1. Operate the drone to fly above the aloe vera field, set the flight altitude to 1.5 meters to 2.5 meters, set the camera deflection angle to be perpendicular to the ground, take a bird's-eye view of the aloe vera field, and shoot and record video data at a horizontal flight speed of less than 1 meter per second; or manually shuttle through the aloe vera field and use a mobile phone to take a straight shot of the surface of the aloe vera leaf, with shooting angles including vertical downward shooting from above and 45 degrees diagonal shooting from the side; you can choose one of the two options, or you can do both;

[0083] S2, pre-process the captured video data, capture a video screenshot every 0.5-1s through the Python program, manually screen out the available pictures and randomly distribute them into training set and test set in a ratio of 9:1;

[0084] S3. Use the Labelme tool to label the aloe objects in the experimental data, convert the output json file into a png image in VOC format for storage. The different color classifications of the pixels in the image represent the healthy leaves, diseased spots and filtered background of the aloe.

[0085] S4. Put the labeled data into the semantic segmentation module, perform deep learning on the improved Deeplab v3+, and use MIoU, MPA and accuracy as evaluation indicators of the algorithm. If the evaluation indicator is less than 0.9, the scheme is improved, the detection model is retrained with the wrong recognition data, and the deep learning results are corrected and stored in the model library, so as to achieve the segmentation detection effect of aloe leaf diseases.

[0086] Furthermore, step S4 is specifically as follows:

[0087] S41. Put the labeled data into the semantic segmentation module, which is based on the improved Deeplab v3+ model.

[0088] S42, DeepLab uses a convolutional neural network (CNN) architecture, which is usually based on improvements to certain standard networks (such as Mobilenet or Xception), and introduces atrous convolution, which enables the network to extract features at different scales without losing resolution.

[0089] S43. During training, DeepLab uses the cross entropy loss function to evaluate the difference between the prediction and the true label. The model will continuously adjust the parameters based on the feedback of the loss.

[0090] S44. After training, the test set is used to evaluate the performance of the model. Indicators such as intersection over union (IoU) and pixel accuracy are usually used to measure the segmentation quality of the model.

[0091] S45. Users can input the required aloe leaf images for testing through the visualization platform. The model can segment the disease spots on the surface of the aloe leaves and realize the classification study of aloe diseases based on the calculation of the proportion of the disease spot area to the total leaf area.

[0092] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0093] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0094] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. Aloe disease grading system based on improved Deeplab v3+, characterized in that: It includes data acquisition module, data preprocessing module, semantic segmentation module, performance estimation module, input module, recognition module, classification module, output module and database; The data acquisition module is used to obtain the leaf data of aloe vera taken from above or from the side; The data preprocessing module preprocesses the collected aloe data video to obtain a picture data set, and marks the aloe leaves and the disease spots on the leaf surfaces, and finally divides the data set; The semantic segmentation module is used to extract features of aloe leaves and surface lesions to obtain aloe leaf recognition and lesion segmentation results; the semantic segmentation module adopts the UAMT network framework, which includes a teacher model Tnet and a student model Snet, and adds constraints on uncertain information in the "prediction consistency" part, so that when calculating the consistency loss, only the results with smaller prediction deviations of the teacher model Tnet and the student model Snet are learned, while the results with larger prediction deviations are suppressed; the teacher model Tnet and the student model Snet have the same structure, and both use the improved Deeplab v3+ network; the UAMT network framework uses the improved Deeplab v3+ network for deep learning in the process of aloe plant detection, and improves the confidence of the network output by evaluating the uncertainty in the prediction process of the teacher network; the lightweight and efficient MobileNetV2 architecture is used to optimize the model backbone network, and a multi-scale attention module is introduced into the backbone network and the CBAM attention mechanism is inserted to improve the efficiency and real-time performance of the semantic segmentation module; The performance estimation module is used to evaluate the detection results, test the MIoU, MPA and accuracy of the detection results, and return to the semantic segmentation module if it is lower than the set threshold, improve the solution, and re-extract features; The input module is used for inputting the picture information required for aloe disease detection into a model; The recognition module is used to recognize and segment the aloe leaves and the diseased spots on their surfaces in the input video or image; The grading module is used to calculate the segmentation results of the recognition module and grade the degree of aloe disease; The output module is used to output and present the disease identification and classification data to the terminal display device, and at the same time output the calculated results in a weighted manner; The database is used to store the characteristics of aloe leaves and surface disease spots.

2. The aloe disease grading system based on the improved Deeplab v3+ according to claim 1, characterized in that: The method of obtaining the aloe leaf data shot from above or from the side is as follows: Use a drone or handheld terminal to take photos to obtain video data of aloe leaves from overhead or side shots.

3. The aloe disease grading system based on improved Deeplab v3+ according to claim 1, characterized in that: The preprocessing is mean normalization, and the feature scaling formula is: In the formula, x′ is the average normalized value, x is the original value; max(x) represents the maximum value of the sample data, and min(x) represents the minimum value of the sample data.

4. The aloe disease grading system based on improved Deeplab v3+ according to claim 1, characterized in that: The improved Deeplab v3+ network includes an encoder and a decoder. The multi-scale attention module MSA introduces a spatial channel attention module CBAM on features of different scales, and then fuses enhanced feature representations of different scales to further enhance the representation capabilities of feature maps of different scales. The Deeplab v3+ network selects the 4th and 8th layers in the backbone network MobilenetV2 as key layers for feature extraction. These two layers correspond to the 1 / 4 and 1 / 16 scales of the input image, respectively, and their initial channel numbers are 24 and 64, respectively, which effectively capture different levels of features from fine texture to high-level semantics. To ensure the consistency of feature maps in the spatial dimension in subsequent processing, the smaller 8th layer feature map is first upsampled by 4 times to match the spatial resolution of the 4th layer feature map, and then the sampled feature maps are respectively input into the CBAM module. Finally, the two feature maps enhanced by the CBAM module are effectively combined together, and the combined feature map contains feature information of different levels and has richer semantic and spatial information.

5. The aloe disease grading system based on the improved Deeplab v3+ according to claim 4, characterized in that: The CBAM module can adaptively adjust the feature maps in the convolutional neural network. When an intermediate feature map is given, the CBAM module calculates attention maps along the channel and spatial dimensions, and then applies these attention maps to the input feature map for adaptive feature improvement.

6. The aloe disease grading system based on improved Deeplab v3+ according to claim 1, characterized in that: The performance estimation module compares the received data with the feature extraction results of the target detection module, and the results include four parts: TP is the positive class judged as the positive class, TN is the negative class judged as the negative class, FN is the positive class judged as the negative class, and FP is the negative class judged as the positive class; MIoU, MPA and accuracy were selected as evaluation indicators of the algorithm; The calculation formula of MIoU is: The calculation formula of MPA is: The accuracy is calculated as follows: The closer all evaluation indicators are to 1, the better the effect.

7. The aloe disease classification method based on improved Deeplab v3+ is characterized by: The steps include: S1, obtaining the leaf data of aloe vera taken from top or side; S2, pre-process the captured video data, capture a video screenshot every 0.5-1s through the Python program, filter out the available pictures and randomly distribute them into training sets and test sets according to the set ratio; S3. Use the Labelme tool to label the aloe objects in the experimental data, convert the output json file into a png image in VOC format for storage. The different color classifications of the pixels in the image represent the healthy leaves, diseased spots and filtered background of the aloe. S4, put the annotated data into the semantic segmentation module, perform deep learning on the improved Deeplab v3+, use MIoU, MPA and accuracy as evaluation indicators of the algorithm, if the evaluation indicator is less than the set threshold, then improve the scheme, retrain the detection model with the wrong recognition data, and store the corrected deep learning results in the model library, so as to achieve the segmentation detection effect of aloe leaf diseases; S5. The user inputs the required aloe leaf images through the visualization platform. The model can segment the lesions on the surface of the aloe leaves and realize the classification study of aloe diseases based on the calculation of the proportion of the lesion area to the total leaf area.

8. The aloe disease grading method based on improved Deeplab v3+ according to claim 7, characterized in that: Step S1 is specifically: Operate the drone to fly above the aloe vera field with the flight altitude set to 1.5 to 2.5 meters and the camera deflection angle set to be perpendicular to the ground to take a bird's-eye view of the aloe vera field, and record video data at a horizontal flight speed of less than 1 meter per second; or manually shuttle through the aloe vera field and use a mobile phone to take direct shots of the surface of the aloe leaves, with shooting angles including vertical downward shooting from above and 45 degrees diagonal shooting from the side.

9. The aloe disease grading method based on improved Deeplab v3+ according to claim 8, characterized in that: In step S4, deep learning is performed on the improved Deeplab v3+, specifically: S41, putting the labeled data into a semantic segmentation module, where the semantic segmentation module is obtained based on an improved Deeplab v3+ model; S42 and DeepLab use the convolutional neural network (CNN) architecture, which is based on the improvement of the selected standard network and introduces dilated convolution, which enables the network to extract features at different scales without losing resolution. S43. During the training process, DeepLab uses the cross entropy loss function to evaluate the difference between the prediction and the true label, and the model will continuously adjust the parameters based on the feedback of the loss. S44. After training, the test set is used to evaluate the performance of the model. Indicators such as intersection over union (IoU) and pixel accuracy are usually used to measure the segmentation quality of the model.

10. The aloe disease grading method based on improved Deeplab v3+ according to claim 7, characterized in that: In step S4, the threshold is 0.9.

Images