CN117911939A - Real-time monitoring and early warning method and system for pine wood nematode disaster based on image segmentation - Google Patents
Real-time monitoring and early warning method and system for pine wood nematode disaster based on image segmentation Download PDFInfo
- Publication number
- CN117911939A CN117911939A CN202311693810.5A CN202311693810A CN117911939A CN 117911939 A CN117911939 A CN 117911939A CN 202311693810 A CN202311693810 A CN 202311693810A CN 117911939 A CN117911939 A CN 117911939A
- Authority
- CN
- China
- Prior art keywords
- model
- early warning
- real
- monitoring
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012544 monitoring process Methods 0.000 title claims abstract description 71
- 241000243771 Bursaphelenchus xylophilus Species 0.000 title claims abstract description 40
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000003709 image segmentation Methods 0.000 title claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 52
- 241000018646 Pinus brutia Species 0.000 claims abstract description 36
- 235000008331 Pinus X rigitaeda Nutrition 0.000 claims abstract description 35
- 235000011613 Pinus brutia Nutrition 0.000 claims abstract description 35
- 238000012360 testing method Methods 0.000 claims abstract description 26
- 238000011156 evaluation Methods 0.000 claims abstract description 24
- 238000007781 pre-processing Methods 0.000 claims abstract description 18
- 238000013135 deep learning Methods 0.000 claims abstract description 17
- 230000006870 function Effects 0.000 claims abstract description 17
- 238000012795 verification Methods 0.000 claims abstract description 7
- 208000015181 infectious disease Diseases 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims abstract description 5
- 201000010099 disease Diseases 0.000 claims description 52
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 52
- 238000013136 deep learning model Methods 0.000 claims description 20
- 241000244206 Nematoda Species 0.000 claims description 19
- 230000004927 fusion Effects 0.000 claims description 16
- 238000002372 labelling Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000000007 visual effect Effects 0.000 claims description 9
- 230000007246 mechanism Effects 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 8
- 230000011218 segmentation Effects 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 7
- 238000004140 cleaning Methods 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 abstract description 5
- 230000002159 abnormal effect Effects 0.000 abstract description 2
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 238000011835 investigation Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 241000243770 Bursaphelenchus Species 0.000 description 2
- 240000007263 Pinus koraiensis Species 0.000 description 2
- 235000011615 Pinus koraiensis Nutrition 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 241000254173 Coleoptera Species 0.000 description 1
- 241000238631 Hexapoda Species 0.000 description 1
- 241000218652 Larix Species 0.000 description 1
- 235000005590 Larix decidua Nutrition 0.000 description 1
- 235000005205 Pinus Nutrition 0.000 description 1
- 241000218602 Pinus <genus> Species 0.000 description 1
- 241000607479 Yersinia pestis Species 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 239000002023 wood Substances 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a real-time monitoring and early warning method and a system for pine wood nematode disaster based on image segmentation, wherein the method comprises the following steps: selecting an unmanned aerial vehicle image positioned in a specific area for acquisition, and preprocessing to obtain a training sample; dividing the generated sample set into a training set, a verification set and a test set, training the training set by adopting a deep learning image processing model, and monitoring the value of a loss function to judge whether the model converges or not; evaluating the model by using the verification set and the test set, and comprehensively evaluating the performance of the model by constructing an evaluation index accuracy, a recall rate and an F1 score; and constructing, monitoring and early warning a user data set. According to the invention, by adopting a high-efficiency deep learning image segmentation-based technology, abnormal areas in the pine forest image acquired by the unmanned aerial vehicle can be accurately segmented, so that the form of the sick pine is judged, the loss and potential infection risk of the sick pine can be timely found and evaluated, and real-time monitoring and early warning service is provided for forestry departments.
Description
Technical Field
The invention relates to the technical field of computer vision and machine learning, in particular to a pine wood nematode disaster real-time monitoring and early warning method and system based on image segmentation.
Background
In the field of forestry, pine wood nematode disease is a disease seriously damaging pine tree growth, and is quick in propagation and strong in concealment, and once outbreaks occur, immeasurable losses are often brought to forest ecology. Although the traditional manual inspection mode can find problems to a certain extent, the traditional manual inspection mode is limited by factors such as personnel, time, region and the like, and comprehensive, systematic and efficient monitoring is difficult to realize.
Pine wood nematodes are very small pests that are difficult to find directly in images of satellite remote sensing or unmanned aerial vehicles. Along with the progress of science and technology, pine wood nematode disease is identified by using pine forest images shot by unmanned aerial vehicles, and whether pine wood nematodes are infected is judged mainly by observing the change of spectral reflectance of pine trees. If a pine is infected with such insects, the tree will gradually yellow and then wilt over a period of several months, and its color will be significantly different from that of a healthy pine, and the diseased tree can be identified by observing this color change.
At present, a method for identifying pine wood nematode disease based on unmanned aerial vehicle images and deep learning technology mainly judges abnormal conditions roughly according to differences of pine tree images, but because of limitations of aerial photographing heights and equipment conditions, detailed shapes and sizes of pine wood are difficult to evaluate specifically, and therefore the degree of disease is difficult to evaluate accurately.
Disclosure of Invention
In view of the above, the invention provides a real-time monitoring and early warning method and system for pine wood nematode disease disasters based on image segmentation, which are used for solving the problem that the pine wood nematode disease disasters are difficult to accurately identify a disease pine area in the prior art.
The specific technical scheme of the invention is as follows:
a real-time monitoring and early warning method for pine wood nematode disease disasters based on image segmentation comprises the following steps:
Step 1, data acquisition and pretreatment: selecting unmanned aerial vehicle images in a preset area for acquisition, and obtaining high-quality training samples through image cleaning, labeling and enhanced preprocessing steps;
step 2, model selection and training: dividing the sample set generated in the step 1 into a training set, a verification set and a test set, training the training set by adopting a deep learning image processing model, monitoring the value of a loss function to judge whether the model converges, and if so, entering the next step;
Step 3, model test and evaluation: evaluating the model by using the verification set and the test set, and comprehensively evaluating the performance of the model by constructing an evaluation index accuracy, a recall rate and an F1 score;
Step 4, constructing, monitoring and early warning a user data set: and (3) carrying out real-time monitoring and early warning by utilizing the user data set according to the optimal model obtained in the step (3).
Further, the step 1 specifically includes: selecting a data acquisition area; unmanned aerial vehicle image acquisition; collecting a real-time image; labeling an image; image preprocessing and dataset generation.
Further, the step 2 specifically includes:
Step 2.1, selecting a plurality of types of deep learning visual models as evaluation objects, wherein the plurality of types of deep learning visual models at least comprise: a full convolution network of semantic segmentation, a multi-scale feature fusion model using pyramid pooling or symmetric encoder-decoder, a model that expands convolution to increase receptive fields, and a model that captures context using a self-attention mechanism instead of multi-scale feature fusion;
and 2.2, training the selected deep learning model by using a Dice loss function.
Further, the Dice loss is defined as
Where |a n b| is the number of pixels in both the predicted value a and the true value B, and |a| and |b| represent the number of pixels in the predicted value a and the true value B, respectively.
Further, the step 3 specifically includes: step 3.1, evaluating the performances of different deep learning models by using the accuracy rate, the recall rate, the Jaccard index and the F1 scoring index; step 3.2, calculating a full convolution network of semantic segmentation, using pyramid pooling or a multi-scale feature fusion model of a symmetrical encoder-decoder, expanding convolution to increase a receptive field model, using a self-attention mechanism to replace multi-scale feature fusion to capture the accuracy rate, recall rate, jaccard index and F1 scoring values of a context model, comparing the scoring values of the models, and selecting the model with the highest value as a candidate model.
Further, the calculation formula of the accuracy is: precision = TP/(tp+fp);
The calculation formula of the recall rate is as follows: recall=tp/(tp+fn);
the Jaccard index is calculated as:
The calculation formula of the F1 scoring value is as follows: f1 =2 ((Precision x Recall)/(precision+recall)); wherein TP indicates that the positive label is correctly predicted as positive, TN indicates that the negative label is correctly predicted as negative, FP indicates that the negative label is incorrectly predicted as positive, and FN indicates that the positive label is incorrectly predicted as negative; j (a, B) represents the Jaccard exponent between the predicted a and the actual value B; the |a n b| is the number of pixels in the predicted value a and the true value B at the same time, and the |a| and the |b| represent the number of pixels in the predicted value a and the true value B, respectively.
Further, step 4 specifically includes: step 4.1, monitoring and evaluating the condition of pine forest nematode disease by using a nematode disease monitoring and evaluating model based on a deep learning model; and 4.2, inputting the user data set into the nematode disease monitoring and evaluating model based on the deep learning model obtained in the step 3, and outputting an evaluating result about the infection condition of the pine forest nematode disease in the area.
A real-time monitoring and early warning system for pine wood nematode disease disasters based on image segmentation, which applies any one of the real-time monitoring and early warning methods for pine wood nematode disease disasters, comprises: the system comprises a data acquisition and preprocessing module, a model selection and training module, a model test and evaluation module and a user data set construction and monitoring and early warning module;
the data acquisition and preprocessing module comprises a data acquisition area selection unit, an unmanned aerial vehicle image acquisition unit, a real image acquisition unit, an image labeling unit and an image preprocessing and data set generation unit;
The model selection and training module comprises a deep learning visual model unit for selecting different types and a training unit for training the selected deep learning model by using a Dice loss function;
The model test and evaluation module comprises a performance unit for evaluating different deep learning models by using an accuracy rate, a recall rate, a Jaccard index and an F1 scoring index, a full convolution network for calculating semantic segmentation, a multi-scale feature fusion model for using pyramid pooling or a symmetrical encoder-decoder, a model for expanding convolution to increase receptive fields, and a model for capturing context by using a self-attention mechanism instead of multi-scale feature fusion, wherein the accuracy rate, the recall rate, the Jaccard index and the F1 scoring index are used, the scoring values of the models are compared, and a model with the highest value is selected as a candidate model unit;
and the user data set construction and monitoring early warning module is used for monitoring and early warning the pine wood nematode disaster in real time, and the user data set is used for carrying out real-time monitoring and early warning according to the optimal model obtained by the model testing and evaluating module.
The invention has the beneficial effects that:
The invention provides a novel pine wood nematode disease monitoring and evaluating method by utilizing a deep learning technology, which can more accurately identify and evaluate the damage condition of the pine wood nematode disease, can reduce the conditions of missed detection and false detection by accurately identifying the pine wood nematode disease area, and improves the monitoring accuracy.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments or the description of the prior art will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a real-time monitoring and early warning method for pine wood nematode disasters based on image segmentation;
Fig. 2 is a schematic diagram showing the effect of the image segmentation of the patient pine according to the present application.
Detailed Description
In order to make the technical problems, technical schemes and beneficial effects to be solved more clear, the invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
The application provides a real-time monitoring and early warning method for a pine wood nematode disaster based on image segmentation, which is shown in figure 1.
Firstly, a scientific and practical data acquisition and preprocessing method is designed aiming at the characteristics of pine forest nematode disease, the method comprises the steps of acquiring image data in pine forest by using an unmanned aerial vehicle, cleaning, labeling, enhancing and the like for the data, and the steps aim at acquiring a high-quality unmanned aerial vehicle image dataset of the pine forest nematode disease and provide a basis for subsequent model training.
Then, performing model selection and training by using the data set generated in the previous step, selecting a deep learning image processing model, constructing a corresponding loss function, continuously monitoring whether the value of the loss function continuously descends in the training process, if the loss function does not descend any more, indicating that the model is converged, and entering the next step of model test and evaluation; if the loss function continues to drop, returning to the step of model selection and training, continuing to train the model, which may involve adjusting parameters of the model, changing the architecture of the model, or using a different optimization algorithm, etc., repeating this process until the loss function no longer drops, and the model converges; once the model converges, the model generated in the previous step is used for model test and evaluation, evaluation indexes including accuracy, recall, F1 score and the like are constructed, so that the performance of the model is comprehensively evaluated, and the optimal model is obtained through multiple experiments and comparison.
And finally, constructing a user data set, utilizing the test result of the last step and the established model to realize the monitoring and evaluation of the pine forest nematode disease based on the deep learning model, specifically, inputting the user data set into an optimal model to obtain prediction results, wherein the prediction results can help to evaluate the performance of the model and determine whether the model can accurately identify the pine forest nematode disease, and simultaneously, a visual interface and an alarm function can be provided so that a user can intuitively check the monitoring result and take corresponding measures.
The specific steps of the invention are as follows:
Step 1, data acquisition and pretreatment: unmanned aerial vehicle images in a specific area are selected for acquisition, and a series of preprocessing steps including image cleaning, labeling, enhancement and the like are carried out to obtain high-quality training samples. The method specifically comprises the following steps:
Step 1.1, selecting a data acquisition area: four areas A1, A2, A3 and A4 located in the southwest of Shu Cheng county of Anhui province are selected as data acquisition areas, the areas belong to mountain terrains, the average elevation is more than 500 meters, and the highest elevation is 1539 meters. The annual average air temperature of the data acquisition area is 15.6 ℃, the annual average precipitation amount is 1200 mm, and the method is suitable for large-scale pine tree planting and belongs to the area with high incidence of pine forest nematode disease.
Step 1.2, unmanned aerial vehicle image acquisition: at 10 months, most of the infected pine was withered, i.e. the infected pine showed a significant difference in color from the other healthy pine, while the other larch and healthy pine still appeared green. Therefore, unmanned aerial vehicle image acquisition is selected to be carried out in 10 months so as to be distinguished from a sick and loose image caused by summer longhorn beetle disasters. The DB-2 fixed wing unmanned aerial vehicle is adopted to collect image data, and the maximum resolution of a lens is 7952 multiplied by 4472 pixels. The unmanned aerial vehicle cruises and sails highly below 700 meters, and the speed is below 100 km/hour. Equidistant shooting is adopted. At least 75% overlap in the direction of flight and at least 50% overlap in the lateral direction. An image resolution of 8cm was finally obtained.
Step 1.3, real image acquisition: and collecting images of different pinus koraiensis through field investigation, and constructing pinus koraiensis identification interpretation knowledge based on unmanned aerial vehicle images. Areas with higher apparent densities are selected for investigation, and when infected pine trees are found, images of the pine trees are taken and their coordinate information is recorded. Through the investigation, 47 pine plants are found out in total, 320 related images are acquired, and the images are used for subsequent data analysis and model training to help monitor and early warn the occurrence of pine wood nematode disease.
Step 1.4, image labeling: to ensure the quality of the training samples, the accuracy of the manual judgment needs to be verified. Therefore, 200 images shot by the unmanned aerial vehicle are selected for testing, and whether the identification of the sick pine is accurate or not is verified according to the results of the field investigation. The results showed that the drone successfully identified about 96% of infected pine, indicating that the judgment accuracy was quite high. Next 2325 images of the drone were collected from four different areas and 35 representative images were selected for annotation. The resolution of these images is high, up to 7952 x 4472 pixels. The number of images selected in each region is different from each other, namely 10 in the A1 region, 5 in the A2 region, 12 in the A3 region and 8 in the A4 region. After labeling (annotation) by using the ENVI remote sensing information processing platform, the positions and the number of 746 infected pine trees are determined. The marked images become important data sources for training models, and help to monitor and identify pine wood nematode diseases more accurately.
Step 1.5, image preprocessing and data set generation: the annotated image is preprocessed and a dataset for training is generated. Firstly, small images are cut out from high-resolution unmanned aerial vehicle images, the size of each small image is 256 pixels multiplied by 256 pixels, and the small images all contain loose pixels; then, the marked large images are rotated 5 times, 10 times and 15 times, and the images after each rotation are randomly cut, so that more training samples are generated, and 36000 samples are obtained in total, wherein the size of each sample is still 256 pixels by 256 pixels; finally, these samples were divided into three parts: 50% for training, 20% for validation, 30% for testing. The purpose of this is to ensure that the model is able to perform stably on different data sets, thus achieving better monitoring results.
Step 2, model selection and training: and (3) selecting a deep learning image processing model to train by utilizing the data set generated in the step (1), and monitoring the value of the loss function to judge whether the model converges or not. The method specifically comprises the following steps:
Step 2.1, selecting a model: different types of deep-learning visual models, such as a full-convolution network (FCNs) for semantic segmentation, a multi-scale feature fusion model (PSPNet) using pyramid pooling or symmetric encoder-decoder, a model that expands convolution to increase receptive fields (e.g., deepLabv 3), and a model that captures context using self-attention mechanisms instead of multi-scale feature fusion (e.g., DANet), etc., are chosen as evaluation targets.
Step 2.2, model training: training the selected deep learning model by using the Dice loss function. In the training process, the training data set in the sample is used for training the model, parameters of the model are continuously adjusted to minimize the loss function, when the loss of the model is no longer reduced, training is stopped, and the model at the moment can be better fitted with the training data set.
The Dice loss is defined based on Dice coefficients for measuring the similarity between model predictive values and true values, and specifically, the Dice loss is defined as
Where |a n b| is the number of pixels in both the predicted value a and the true value B, and |a| and |b| represent the number of pixels in the predicted value a and the true value B, respectively.
Step 3, model test and evaluation: and evaluating the model by using the verification set and the test set, and comprehensively evaluating the performance of the model by constructing evaluation indexes such as accuracy, recall, F1 score and the like. The method specifically comprises the following steps:
step 3.1, test index structure: the performance of different deep learning models was evaluated using the indices of accuracy, recall, jaccard index, and F1 score. The specific calculation is as follows:
(1) The calculation formula of accuracy (Precision) is: precision = TP/(tp+fp)
(2) The calculation formula of the Recall ratio (Recall) is: recall=tp/(tp+fn)
(3)
(4) The calculation formula of the F1 scoring value is as follows: f1 =2× ((precision×recall)/(precision+recall))
The Precision is the accuracy, which indicates that the model predicts the correct proportion of the pinus korotus; TP represents true positive (positive label correctly predicted as positive), TN represents true negative (negative label correctly predicted as negative), FP represents false positive (negative label incorrectly predicted as positive), FN represents false negative (positive label incorrectly predicted as negative); recall is Recall rate, and the ability of the evaluation model to find all pints; for Precision (Precision) and Recall (Recall), the optimal value is 1, and the worst value is 0; the Jaccard index, also called Jaccard similarity coefficient, is defined as the size of the intersection divided by the size of the union of the two sets of labels, and has a value of [0,1], with a value of 0 indicating no overlap and 1 indicating complete overlap; j (a, B) represents the Jaccard exponent between the predicted a and the actual value B; f1 score, also known as the Dice similarity coefficient, is the harmonic mean of Precision (Precision) and Recall (Recall), ranging from 0,1 for perfect Precision and Recall, 0 for zero Precision or Recall.
Step 3.2, evaluation of the optimal model: calculating FCNs, PSPNet, deepLabv, DANet and other models, wherein the accuracy rate refers to the correct proportion of model prediction, the recall rate refers to the capacity of the model to find all pints, the Jaccard index refers to the similarity between the model prediction result and the true value, and the F1 score refers to the harmonic mean of the accuracy rate and the recall rate. And comparing the scoring values of the models, and selecting the model with the highest value as a candidate model. In this embodiment DeepLabv performs better than other models in terms of various indices, and is therefore selected as the final method of nematode disease monitoring and assessment based on the deep learning model.
Step 4, constructing, monitoring and early warning a user data set: and (3) carrying out real-time monitoring and early warning by utilizing the user data set according to the optimal model obtained in the step (3). The method specifically comprises the following steps:
Step 4.1, monitoring and evaluating conditions of pine forest nematode disease by using a nematode disease monitoring and evaluating model based on a deep learning model: in order to obtain the user data set, the method according to steps 1.1, 1.2 and 1.5 is required to acquire an image of the target region and to crop. These steps include selecting an appropriate image acquisition device, determining the area and object to acquire, and cropping and preprocessing the image using image processing software. Through these operations, a user data set suitable for model training and testing can be obtained.
Step 4.2, inputting the user data set into the nematode disease monitoring and evaluating model based on the deep learning model obtained in the step 3, and outputting an evaluating result about the infection condition of the pine forest nematode disease in the area: the input of the model includes various feature extraction and image processing methods to extract information useful for pine wood nematode disease monitoring and assessment. Finally, the model outputs an evaluation result about infection of the regional pine forest nematode disease. These results may include information on the extent, extent and trend of infection, providing decision support for forestry protection and management.
The application also provides a real-time monitoring and early warning system for pine wood nematode disease disasters based on image segmentation, which comprises the following steps: the system comprises a data acquisition and preprocessing module, a model selection and training module, a model test and evaluation module and a user data set construction and monitoring and early warning module.
The data acquisition and preprocessing module comprises a data acquisition area selection unit, an unmanned aerial vehicle image acquisition unit, a real image acquisition unit, an image labeling unit and an image preprocessing and data set generation unit.
The model selection and training module comprises a deep learning visual model unit for selecting different types and a training unit for training the selected deep learning model by using a Dice loss function.
The model test and evaluation module includes evaluating performance units of different deep learning models using accuracy, recall, jaccard index, and F1 score, computing a full convolution network of semantic partitions, a multi-scale feature fusion model using pyramid pooling or symmetric encoder-decoder, expanding a convolution to increase receptive fields, and capturing accuracy, recall, jaccard index, and F1 score values for models of context using a self-attention mechanism instead of multi-scale feature fusion, comparing score values for each model, and selecting a model with the highest value as a candidate model unit.
And the user data set construction and monitoring early warning module is used for monitoring and early warning the pine wood nematode disaster in real time, and the user data set is used for carrying out real-time monitoring and early warning according to the optimal model obtained by the model testing and evaluating module.
The schematic diagram of the effect of the division of the sick and loose image is shown in fig. 2, so that the sick and loose contour can be more accurately divided from the aerial image of the unmanned aerial vehicle according to the method provided by the application; the gray level difference of the exposed rock of the mountain and the like does not influence the disease-loosening identification and segmentation capability of the method provided by the application.
The application has the beneficial effects that:
(1) The monitoring accuracy is improved: the invention designs a novel pine wood nematode disease monitoring and evaluating method and system by utilizing a deep learning technology, which can more accurately identify and evaluate the hazard condition of the pine wood nematode disease, can reduce the conditions of missed detection and false detection by accurately identifying the pine wood nematode disease area, and improves the monitoring accuracy;
(2) Improving the quality of the data set: the invention designs a new data set construction method for pine wood nematode disease monitoring, and constructs an unmanned aerial vehicle image specialized data set applied to pine wood nematode disease monitoring, which can better meet the requirement of pine wood nematode disease monitoring, improve the accuracy of model training, and further improve the accuracy and reliability of monitoring results;
(3) And (3) training an optimization model: the invention designs a novel model evaluation method for pine wood nematode disease monitoring, which is beneficial to optimizing a model training process, and can timely find and correct problems in model training, and improve model performance, thereby improving the accuracy of monitoring and evaluation results.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Claims (8)
1. A real-time monitoring and early warning method for pine wood nematode disease disasters based on image segmentation is characterized by comprising the following steps:
Step 1, data acquisition and pretreatment: selecting unmanned aerial vehicle images in a preset area for acquisition, and obtaining high-quality training samples through image cleaning, labeling and enhanced preprocessing steps;
step 2, model selection and training: dividing the sample set generated in the step 1 into a training set, a verification set and a test set, training the training set by adopting a deep learning image processing model, monitoring the value of a loss function to judge whether the model converges, and if so, entering the next step;
Step 3, model test and evaluation: evaluating the model by using the verification set and the test set, and comprehensively evaluating the performance of the model by constructing an evaluation index accuracy, a recall rate and an F1 score;
Step 4, constructing, monitoring and early warning a user data set: and (3) carrying out real-time monitoring and early warning by utilizing the user data set according to the optimal model obtained in the step (3).
2. The real-time monitoring and early warning method for pine wood nematode disease disasters according to claim 1, wherein the step 1 specifically comprises the following steps: selecting a data acquisition area; unmanned aerial vehicle image acquisition; collecting a real-time image; labeling an image; image preprocessing and dataset generation.
3. The real-time monitoring and early warning method for pine wood nematode disease disasters according to claim 1, wherein the step 2 specifically comprises the following steps:
Step 2.1, selecting a plurality of types of deep learning visual models as evaluation objects, wherein the plurality of types of deep learning visual models at least comprise: a full convolution network of semantic segmentation, a multi-scale feature fusion model using pyramid pooling or symmetric encoder-decoder, a model that expands convolution to increase receptive fields, and a model that captures context using a self-attention mechanism instead of multi-scale feature fusion;
and 2.2, training the selected deep learning model by using a Dice loss function.
4. The real-time monitoring and early warning method for pine wood nematode disease disasters according to claim 3, wherein the Dice loss is defined as
Where |a n b| is the number of pixels in both the predicted value a and the true value B, and |a| and |b| represent the number of pixels in the predicted value a and the true value B, respectively.
5. The real-time monitoring and early warning method for pine wood nematode disease disasters according to claim 1, wherein the step 3 specifically comprises the following steps: step 3.1, evaluating the performances of different deep learning models by using the accuracy rate, the recall rate, the Jaccard index and the F1 scoring index; step 3.2, calculating a full convolution network of semantic segmentation, using pyramid pooling or a multi-scale feature fusion model of a symmetrical encoder-decoder, expanding convolution to increase a receptive field model, using a self-attention mechanism to replace multi-scale feature fusion to capture the accuracy rate, recall rate, jaccard index and F1 scoring values of a context model, comparing the scoring values of the models, and selecting the model with the highest value as a candidate model.
6. The real-time monitoring and early warning method for pine wood nematode disease disasters according to claim 5, which is characterized in that,
The calculation formula of the accuracy is: precision = TP/(tp+fp);
The calculation formula of the recall rate is as follows: recall=tp/(tp+fn);
the Jaccard index is calculated as:
The calculation formula of the F1 scoring value is as follows: f1 =2 ((Precision x Recall)/(precision+recall)); wherein TP indicates that the positive label is correctly predicted as positive, TN indicates that the negative label is correctly predicted as negative, FP indicates that the negative label is incorrectly predicted as positive, and FN indicates that the positive label is incorrectly predicted as negative; j (a, B) represents the Jaccard exponent between the predicted a and the actual value B; the |a n b| is the number of pixels in the predicted value a and the true value B at the same time, and the |a| and the |b| represent the number of pixels in the predicted value a and the true value B, respectively.
7. The real-time monitoring and early warning method for pine wood nematode disease disasters according to claim 5, wherein the step 4 specifically comprises the following steps: step 4.1, monitoring and evaluating the condition of pine forest nematode disease by using a nematode disease monitoring and evaluating model based on a deep learning model; and 4.2, inputting the user data set into the nematode disease monitoring and evaluating model based on the deep learning model obtained in the step 3, and outputting an evaluating result about the infection condition of the pine forest nematode disease in the area.
8. The real-time monitoring and early warning system for pine wood nematode disasters based on image segmentation, which is applied to the real-time monitoring and early warning method for pine wood nematode disasters according to any one of claims 1-8, is characterized in that the system comprises: the system comprises a data acquisition and preprocessing module, a model selection and training module, a model test and evaluation module and a user data set construction and monitoring and early warning module;
the data acquisition and preprocessing module comprises a data acquisition area selection unit, an unmanned aerial vehicle image acquisition unit, a real image acquisition unit, an image labeling unit and an image preprocessing and data set generation unit;
the model selection and training module comprises a deep learning visual model unit for selecting different types and a training unit for training the selected deep learning model by using a Dice loss function;
the model test and evaluation module comprises a performance unit for evaluating different deep learning models by using an accuracy rate, a recall rate, a Jaccard index and an F1 scoring index, a full convolution network for calculating semantic segmentation, a multi-scale feature fusion model for increasing a receptive field by using pyramid pooling or a symmetrical encoder-decoder, a model for increasing the receptive field by expanding convolution and a model for capturing a context by using a self-attention mechanism instead of multi-scale feature fusion, wherein the accuracy rate, the recall rate, the Jaccard index and the F1 scoring index are used, the scoring values of the models are compared, and a model with the highest value is selected as a candidate model unit;
the user data set construction and monitoring early warning module is used for monitoring and early warning pine wood nematode disasters in real time, and the user data set is used for carrying out real-time monitoring and early warning according to the optimal model obtained by the model testing and evaluating module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311693810.5A CN117911939A (en) | 2023-12-07 | 2023-12-07 | Real-time monitoring and early warning method and system for pine wood nematode disaster based on image segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311693810.5A CN117911939A (en) | 2023-12-07 | 2023-12-07 | Real-time monitoring and early warning method and system for pine wood nematode disaster based on image segmentation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117911939A true CN117911939A (en) | 2024-04-19 |
Family
ID=90696639
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311693810.5A Pending CN117911939A (en) | 2023-12-07 | 2023-12-07 | Real-time monitoring and early warning method and system for pine wood nematode disaster based on image segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117911939A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118115903A (en) * | 2024-04-30 | 2024-05-31 | 安徽大学 | Intelligent identification processing system for pine wood nematode disease based on AI technology |
-
2023
- 2023-12-07 CN CN202311693810.5A patent/CN117911939A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118115903A (en) * | 2024-04-30 | 2024-05-31 | 安徽大学 | Intelligent identification processing system for pine wood nematode disease based on AI technology |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110378869B (en) | Steel rail fastener abnormity detection method with automatic sample marking function | |
| CN111178197B (en) | Mass R-CNN and Soft-NMS fusion based group-fed adherent pig example segmentation method | |
| CN109166094A (en) | A kind of insulator breakdown positioning identifying method based on deep learning | |
| CN111209434A (en) | Substation equipment inspection system and method based on multi-source heterogeneous data fusion | |
| CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
| CN111681209B (en) | Intelligent detection system for cleavage state of blastomere | |
| CN112347943A (en) | Anchor optimization safety helmet detection method based on YOLOV4 | |
| CN114565826A (en) | Agricultural pest and disease identification and diagnosis method, system and device | |
| CN114694178A (en) | Method and system for monitoring safety helmet in power operation based on fast-RCNN algorithm | |
| Mohammadpoor et al. | An intelligent technique for grape fanleaf virus detection | |
| CN114462475A (en) | Unsupervised machine abnormal sound detection method and unsupervised machine abnormal sound detection device based on single classification algorithm | |
| CN113378831B (en) | Mouse embryo organ identification and scoring method and system | |
| CN117557840B (en) | Fundus lesion grading method based on small sample learning | |
| CN113674225A (en) | Power equipment fault detection method based on convolutional neural network | |
| CN116310913B (en) | Natural resource investigation monitoring method and device based on unmanned aerial vehicle measurement technology | |
| CN112861881A (en) | Honeycomb lung recognition method based on improved MobileNet model | |
| CN116863388A (en) | Sperm motility determining method and system based on neural network | |
| CN117911939A (en) | Real-time monitoring and early warning method and system for pine wood nematode disaster based on image segmentation | |
| CN116958643A (en) | Air-borne pollen sensitized plant intelligent identification method based on YOLO network | |
| CN116503612A (en) | Fan blade damage identification method and system based on multitasking association | |
| CN115311601A (en) | Fire detection analysis method based on video analysis technology | |
| CN114581705A (en) | Fruit ripening detection method and system based on YOLOv4 model and convolutional neural network | |
| CN116152537A (en) | 15 forestry pest identification algorithms based on Yolov5s | |
| CN111652102A (en) | Power transmission channel target object identification method and system | |
| Luo et al. | Peach tree detection for weeding robot based on Faster-RCNN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |