CN115063700A - Detection method based on small-scale pine wood nematode disease tree - Google Patents
Detection method based on small-scale pine wood nematode disease tree Download PDFInfo
- Publication number
- CN115063700A CN115063700A CN202210617043.9A CN202210617043A CN115063700A CN 115063700 A CN115063700 A CN 115063700A CN 202210617043 A CN202210617043 A CN 202210617043A CN 115063700 A CN115063700 A CN 115063700A
- Authority
- CN
- China
- Prior art keywords
- feature map
- feature
- pine wood
- wood nematode
- tree
- 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
- 201000010099 disease Diseases 0.000 title claims abstract description 104
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 title claims abstract description 104
- 241000243771 Bursaphelenchus xylophilus Species 0.000 title claims abstract description 70
- 238000001514 detection method Methods 0.000 title claims abstract description 30
- 230000004927 fusion Effects 0.000 claims abstract description 37
- 238000000605 extraction Methods 0.000 claims abstract description 18
- 238000012545 processing Methods 0.000 claims abstract description 18
- 238000009826 distribution Methods 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims description 25
- 238000000034 method Methods 0.000 claims description 17
- 235000008331 Pinus X rigitaeda Nutrition 0.000 claims description 13
- 235000011613 Pinus brutia Nutrition 0.000 claims description 13
- 241000018646 Pinus brutia Species 0.000 claims description 13
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 12
- 230000000694 effects Effects 0.000 claims description 9
- 238000013508 migration Methods 0.000 claims description 6
- 230000005012 migration Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 3
- 238000011835 investigation Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 241000243770 Bursaphelenchus Species 0.000 description 2
- 241000607479 Yersinia pestis Species 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 241000196324 Embryophyta Species 0.000 description 1
- 240000005926 Hamelia patens Species 0.000 description 1
- 241000238631 Hexapoda Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 244000062645 predators Species 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/17—Terrestrial scenes taken from planes or by drones
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Remote Sensing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Image Processing (AREA)
Abstract
A detection method based on small-scale pine wood nematode disease trees comprises the following steps: step 1: acquiring a remote sensing image by using an unmanned aerial vehicle for aerial photography, marking the pine wood nematode disease tree, and making into a data set; step 2: inputting the data set into a bottom layer feature batch processing fusion and multiple feature multiplexing network model, performing feature extraction to obtain an identification model, and identifying the pine wood nematode disease tree; and step 3: vectorizing the identification result of the pine wood nematode disease tree to obtain a longitude and latitude coordinate file of the pine wood nematode disease tree; and 4, step 4: and uploading the longitude and latitude coordinate information of the pine wood nematode disease tree to a pine wood nematode disease tree supervision platform, checking the geographical distribution condition of the disease tree through the supervision platform, and manually felling. The invention aims to solve the problem that the detection of small-scale pine wood nematode disease trees is missed under the complex background of remote sensing images, and provides a pine wood nematode disease tree detection method based on bottom-layer feature batch processing fusion and a multiple-feature multiplexing network model.
Description
Technical Field
The invention relates to the technical field of pine wood nematode disease tree detection, in particular to a detection method based on small-scale pine wood nematode disease trees.
Background
Forest diseases and insect pests are the predators for forest health and forestry production. Pine wood nematodes are one of the foreign invasive species with great harm to pine trees, pine needle leaves are yellow brown or red brown after the pine wood nematodes are infected, the whole diseased tree can be withered and die within 6 months, and the pine wood nematode disease control agent has the advantages of high propagation speed, wide propagation path, wide propagation range and difficulty in control, and is an important pest for destroying large pine forest. Therefore, timely discovery of the pine wood nematode disease tree is of great importance.
The existing monitoring means for the pine wilt disease mainly comprise ground investigation, satellite remote sensing monitoring, multispectral unmanned aerial vehicle images and unmanned aerial vehicle remote sensing monitoring. The traditional investigation method for the pine wilt disease has the advantages of low efficiency, high cost and easy omission due to the utilization of manual general investigation; the satellite remote sensing space resolution is low and the timeliness is poor; multispectral unmanned aerial vehicle image acquisition inefficiency, resolution ratio is high. Obtaining panchromatic wave band images through an unmanned aerial vehicle is the most effective at present, the panchromatic wave band images of the unmanned aerial vehicle can obtain images with higher spatial resolution, meanwhile, the efficiency of obtaining the images can be guaranteed, and by extracting the characteristics of colors, shapes, textures and the like of single plants of sick trees on the images with high spatial resolution, higher detection precision can be obtained through a deep learning mode, and the requirement of treatment is met.
Since the time of 2012, after image network large-scale visual recognition challenges image classification competition, AlexNet proposed by Hinton et al develops rapidly in deep learning, and is applied more and more widely in the field of computer vision, and more scholars apply deep learning to target detection in remote sensing scenes. Wang et al propose an SSD improvement algorithm in combination with a Feature Pyramid (FPN) for remote sensing image small target detection, and the algorithm improves the speed and precision of detection to a certain extent. Yaoqun power et al propose a multiscale network remote sensing target detection framework-MSCNN, this algorithm is better to the large-scale target result of multiscale remote sensing, but to the small-scale target detection effect under the multiscale is relatively poor.
In order to realize the efficient detection of the remote sensing small target under the complex background, the patent provides a detection method of the pine wood nematode disease tree of a bottom layer characteristic batch processing fusion and multiple characteristic multiplexing network model. The method considers and redesigns a feature fusion module, fuses a shallow feature map mainly responsible for small target detection and a deep feature map with rich semantic information, adopts a batch processing modular feature fusion mode to enhance and multiplex the shallow feature information, improves the feature extraction capability of small-scale targets, adopts the idea of transfer learning, firstly performs model training on relatively small-scale disease tree samples under low resolution, and then takes a model obtained by training as a higher resolution 1: 500, thereby improving the detection capability of the network model to small-scale target disease trees.
Disclosure of Invention
The invention aims to solve the problem that the detection of small-scale pine wood nematode disease trees is missed under the complex background of remote sensing images, and provides a pine wood nematode disease tree detection method based on bottom-layer feature batch processing fusion and a multiple-feature multiplexing network model.
A detection method based on small-scale pine wood nematode disease trees specifically comprises the following steps:
step 1: acquiring an image, marking the pine wood nematode disease tree, and making into a data set;
step 2: inputting the data set into a bottom layer feature batch processing fusion and multiple feature multiplexing network model, performing feature extraction to obtain an identification model, and identifying the pine wood nematode disease tree;
and 3, step 3: vectorizing the identification result of the pine wood nematode disease tree to obtain longitude and latitude coordinate information of the pine wood nematode disease tree;
and 4, step 4: and uploading the longitude and latitude coordinate information of the pine wood nematode disease tree to a pine wood nematode disease tree supervision platform, checking the geographical distribution condition of the disease tree through the supervision platform, and manually felling.
In step 2, inputting the data set into a bottom layer feature batch processing fusion and multiple feature multiplexing network model, performing feature extraction to obtain an identification model, and specifically adopting the following steps:
2-1: selecting and preprocessing an unmanned aerial vehicle image sample set;
2-2: classifying the data sets according to the disease tree pictures with different scales, and separately carrying out migration training on the network model;
2-3: and inputting the data set into a bottom layer feature batch processing fusion and multiple feature multiplexing network model, extracting multiple feature information to obtain a pine wood nematode disease tree identification model, and identifying the pine wood nematode disease tree.
In step 2-2, the data sets are classified according to different scales, and the network model is separately migrated and trained, specifically comprising the following substeps:
2-2-1: in a data set sample with the scale value A, the scale of the disease tree is relatively small, so that the characteristic extraction of the small-scale disease tree by the model is facilitated, the data set with the scale value A is firstly input into a network model, and a pine wood nematode disease tree recognition model is obtained through training;
2-2-2: and then inputting the data set sample with the size of B into a network model, and performing migration training by taking the recognition model obtained by training with the size of A as a pre-training model for training the data set sample with the size of B to obtain the final pine wood nematode disease tree recognition model.
In step 2-3, constructing a bottom-layer feature batch processing fusion and multiple feature multiplexing network model, specifically comprising the following substeps:
2-3-1: inputting an image to be detected into a feature extraction backbone network, and extracting features from a shallow layer to a deep layer in a bottom-up mode, wherein the features are as follows:
performing down-sampling operation on the input graph through convolution to obtain a feature graph L1;
performing maximum pooling downsampling operation on the feature map L1 through convolution, and sequentially performing a plurality of convolution residual block operations to obtain a feature map L2; carrying out a plurality of convolution residual block operations on the feature map L2 to obtain a feature map L3; performing a plurality of convolution residual block operations on the feature map L3 to obtain a feature map L4; and subjecting the feature map L4 to a plurality of convolution residual block operations to obtain a feature map L5.
2-3-2: the method comprises the following steps of carrying out batch processing fusion on bottom layer features and multiplexing deep layer features to enhance the feature extraction capability of a model on the pine wood nematode disease trees with different scales, and then identifying the pine wood nematode disease trees, wherein the method comprises the following specific steps:
performing convolution operation on the bottom layer feature map L1, the feature map L2 and the feature map L3 to enable the channel numbers of the bottom layer feature map L1, the feature map L2 and the feature map L3 to be the same, and then performing feature fusion to obtain a feature map F2;
performing convolution operation on the feature map F2 by using a convolution kernel of 3 × 3 and convolution with the step length of 1, eliminating the aliasing effect after fusion, and obtaining a final predicted feature map Ps1 for detecting the small-scale disease tree;
performing convolution operation on the deep feature maps L4 and L5 to enable the channel numbers of the feature maps L4 and L5 to be the same as the channel number of the feature map L3; then carrying out feature fusion to obtain a feature map F3;
performing convolution operation on the feature map F3 by using a convolution kernel of 3 × 3 and convolution with the step length of 1, eliminating the aliasing effect after fusion, and obtaining a final predicted feature map Ps2 for detecting a small-scale disease tree which is not detected by the predicted feature map Ps 1;
fusing the feature map L4 with the feature map F2 subjected to the downsampling operation and the feature map L5 subjected to the upsampling operation to obtain a feature map F4;
performing convolution operation on the feature map F4 by using a convolution kernel of 3 x 3 and convolution with the step length of 1, eliminating the aliasing effect after fusion, and obtaining a prediction feature map Pm for detecting the mesoscale disease tree;
fusing the feature map L5 with the feature map F2 subjected to downsampling operation to obtain a feature map F5, and performing convolution operation on the feature map F5 to obtain a predicted feature map Pl for detecting the large-scale disease tree;
in step 2, the loss function adopted in the bottom layer feature batch fusion and multiple feature multiplexing network model is a focus loss function, which is specifically as follows:
the meaning of the individual parameters is as follows: y is the true tag value (positive sample value is 1, negative sample value is 0); y' is the prediction class probability given by the model; gamma is a hyper-parameter, gamma is greater than 0, so that the loss of samples which are easy to classify is reduced, the model is more concerned with difficult and misclassified samples, the reduction rate of the weight of a simple sample can be adjusted by changing the size of gamma, the gamma is a cross entropy loss function when the gamma is 0, the influence of an adjustment factor is increased when the gamma is increased, the gamma can automatically reduce the weight of the contribution of the simple sample (the sample with more definite class) to loss during training, and the model is more concerned with the difficult samples (the samples with less distinguishable classes) rapidly, wherein the gamma is optimal when the gamma takes a value of 2; alpha is a balance factor over-parameter used for balancing the proportion unevenness of the positive and negative samples, and the value is 0.25.
In step 4, the following substeps are specifically included:
4-1: importing the longitude and latitude coordinate information of the identified diseased tree into a pine wood nematode disease tree supervision platform;
4-2: the geographical distribution condition of the diseased trees can be checked on the pine wood nematode disease tree supervision platform, the diseased trees are found according to the diseased tree navigation information of the supervision platform, and the diseased pines are surveyed and manually felled.
Compared with the prior art, the invention has the following technical effects:
the invention relates to a detection method based on a small-scale pine wood nematode disease tree, which only detects the whole pine wood nematode disease tree at present, does not research on the detection of the small-scale disease tree, and has the condition of missing detection of the small-scale pine wood nematode disease tree under the complex background of a remote sensing image, so that the detection method based on a network model of bottom-layer feature batch processing fusion and multiple feature multiplexing is provided. According to the method, a bottom-layer feature batch processing fusion module is designed, the shallow feature map mainly responsible for small target detection is subjected to adjacent feature fusion, deep features with rich semantic information are fused with the shallow features, and the feature extraction capability of the model on the small target is improved.
Drawings
The invention is further illustrated with reference to the following figures and examples:
FIG. 1 is a flow chart of the present invention;
fig. 2 is a diagram showing a network structure in the present invention.
The specific implementation mode is as follows:
as shown in fig. 1, a method for detecting a pine wood nematode disease tree based on small scale comprises the following steps:
step 1: acquiring a remote sensing image by using an unmanned aerial vehicle for aerial photography, marking the pine wood nematode disease tree by using a labellimg tool, and making into a data set;
step 2: inputting the data set into a bottom layer feature batch processing fusion and multiple feature multiplexing network model, performing feature extraction to obtain an identification model, and identifying the pine wood nematode disease tree;
and step 3: vectorizing the identification result of the pine wood nematode disease tree to obtain a longitude and latitude coordinate file of the pine wood nematode disease tree;
and 4, step 4: and uploading the longitude and latitude coordinate information of the pine wood nematode disease tree to a pine wood nematode disease tree supervision platform, checking the geographical distribution condition of the disease tree through the supervision platform, and manually felling.
In step 1, an unmanned aerial vehicle is used for obtaining a remote sensing image through aerial photography, a labelimg tool is used for marking a disease tree in the remote sensing image, a data set is manufactured, and the following steps are specifically adopted:
(1) clear pine 1 was obtained using a Da Jiang unmanned aerial vehicle during the peak period of pine wood nematode disease tree incidence of 9 months: 1000 and 1: 500, 1: cutting a 500 HD pine image into 1000 × 1000, and acquiring a 1: cutting 1000 pine images into 1500 × 1500 sizes, and labeling diseased trees in the pictures by using a labelimg tool;
(2) labeling samples according to 1: 500 and 1: 1000 into two data sets of different dimensions.
As shown in fig. 2, in step 2, inputting the data set into the bottom-layer feature batch fusion and multiple-feature multiplexing network model, and performing feature extraction to obtain the recognition model, specifically including the following steps:
(1) performing downsampling operation on the input graph through convolution of 7 × 7 with the step size of 2 to obtain a feature graph L1; performing maximum pooling downsampling operation on the feature map L1 through convolution with step size of 2 by 3, and sequentially performing convolution residual blocks of 3 by 1, 3 by 3 and 1 by 1 to obtain a feature map L2; carrying out 4 convolution residual block operations on the feature map L2 to obtain a feature map L3; carrying out 6 convolution residual block operations on the feature map L3 to obtain a feature map L4; and then, the feature map L4 is subjected to 3 convolution residual block operations to obtain a feature map L5.
Performing 1 × 1 convolution operation on the bottom layer feature map L1, the feature map L2 and the feature map L3 to enable the number of channels to be the same, then performing feature fusion to obtain a feature map F2, performing 3 × 3 convolution operation on the feature map F2 to obtain a final predicted feature map Ps1, and taking charge of detecting the small-scale disease tree;
carrying out 1-to-1 convolution on the deep feature maps L4 and L5 to enable the channel numbers of the feature maps L4 and L5 to be the same as that of the feature map L3, then carrying out feature fusion to obtain a feature map F3, carrying out 3-to-3 convolution on the feature map F3 to obtain a final predicted feature map Ps2, and detecting the missing small-scale disease tree of the feature map Ps 1;
fusing the feature map L4 with a feature map F2 which is subjected to down-sampling operation and is fused with the bottom layer feature map L1, the feature map L2 and the feature map L3 and a feature map L5 which is subjected to up-sampling operation to obtain a feature map F4, and performing 3 × 3 convolution operation on the feature map F4 to obtain a predicted feature map Pm which is responsible for detecting the medium-scale disease tree;
fusing the feature map L5 with the feature map F2 subjected to the downsampling operation to obtain a feature map F5, and performing 3-by-3 convolution operation on the feature map F5 to obtain a predicted feature map Pl which is responsible for detecting the large-scale disease tree;
the traditional small-scale target prediction method is to output a prediction result of a small-scale target at a third layer, but the convolution result of the first layer and the convolution result of the second layer are also helpful for small-scale feature fusion, and the small-scale target information of the second layer is more, so that the features of the first layer and the third layer are fused and added to the second layer to predict the small-scale target, and the small-scale feature extraction capability is enhanced.
On the fourth layer of the deep characteristic diagram, the traditional characteristic fusion mode is that the convolution layer is directly fused with the characteristic diagram sampled on the fifth layer, and the geometric information of the bottom layer is lost;
on the fifth layer of the deep characteristic diagram, the conventional mode is to directly obtain a prediction diagram on the fifth layer of the characteristic diagram through convolution operation, and the shallow characteristic is also fused into the fifth layer of the characteristic diagram, so that the characteristic extraction capability of the model is improved;
(2) inputting a data set sample into a constructed network model, repeating training for multiple times, removing a sample with a wrong identification each time, inputting the sample as a negative sample into the network model, continuing training, in the identification process, due to the characteristic that the mountain forest terrain is complex, other red head, roof, bare land and trees similar to pine trees which interfere with the pine wood nematode disease tree detection exist, removing the sample with the wrong identification each time in the identification process, marking the sample as the negative sample again, putting the sample into a data set, and then training until a stable model identification rate with higher relative precision exists.
(3) In the following steps of 1: in 1000 data samples, the size of the disease tree is relatively small, which is beneficial to small-size disease tree identification, and the method comprises the following steps of 1: 1000, inputting the data set into a network model, and training to obtain the pine wood nematode disease tree recognition model. And then, mixing the mixture of 1: 1000 training the resulting model as a 1: and (5) training a pre-training model trained by 500 data samples to obtain a final recognition model.
(4) The loss function used in the network model is the focus loss, which is specified as follows:
the focus loss is modified on the basis of a cross-entropy loss function, and the two-class cross-entropy loss is as follows:
wherein y is a real label value (the real sample value is 1, and the negative sample value is 0), y' is a prediction class probability given by the model, and the larger the cross entropy loss is, the smaller the loss is, for the positive sample. For negative samples, the smaller the output probability, the smaller the penalty. The loss function at this point is slow in the iterative process of a large number of simple samples and may not be optimized to be optimal.
The focus loss is added with a factor on the original basis, as follows:
where γ is a hyper-parameter, γ >0 allows for reduced loss of easily sorted samples, making the model more concerned with difficult, misclassified samples. Changing the magnitude of gamma also adjusts the rate at which simple sample weights are reduced, as a function of cross-entropy loss when gamma is 0, and the effect of the adjustment factor increases as gamma increases. The gamma automatically reduces the weight of the contribution of the simple samples (the samples with definite categories) to the loss during training, and quickly makes the model focus more on the difficult samples (the samples with the categories which are not easy to distinguish), wherein the gamma is optimal when the gamma takes 2.
In addition, a balance factor α is added to balance the non-uniform ratio of the positive and negative samples themselves, as follows:
vectorizing the identification result of the pine wood nematode disease tree in the steps 3 and 4 to obtain a longitude and latitude coordinate file of the pine wood nematode disease tree; uploading longitude and latitude coordinate information of the pine wood nematode disease tree to a pine wood nematode disease tree supervision platform, checking the geographical distribution condition of the disease tree through the supervision platform, and manually felling the disease tree, wherein the method specifically comprises the following steps:
(1) inputting the pine forest remote sensing image which has coordinate information and needs to be identified into a detection model to obtain an identification result of the pine wood nematode disease tree and position information of the disease tree, and converting the identification result of the pine wood nematode disease tree into longitude and latitude coordinate information. Importing the longitude and latitude information of the pine wood nematode disease tree into ArcGis software, and checking the distribution condition of the disease tree for statistical analysis;
(2) importing the longitude and latitude coordinate information of the identified diseased tree into a pine wood nematode disease tree supervision platform;
(3) the pine wood nematode diseased tree supervision platform can check the geographical distribution condition of diseased trees, find diseased trees according to the diseased tree navigation information of the supervision platform, and investigate and clear diseased pines.
The invention designs the bottom layer characteristic batch processing modularization fusion and the deep layer characteristic multiplexing fusion, fuses the shallow layer characteristic graph mainly responsible for small target detection and the deep layer characteristic graph with rich semantic information, and adopts the combination mode of the bottom layer characteristic adjacent batch processing fusion and the deep layer characteristic multiplexing fusion to enhance the shallow layer information and improve the extraction capability of the model to the small target characteristic.
Meanwhile, the invention adopts a migration experiment mode, firstly trains the small-scale sick tree under low resolution, and then migrates to the data set under higher resolution for training, so that the model learns the characteristics of the small-scale sick tree firstly, and the detection capability of the network model on the small-scale sick tree is improved.
Claims (6)
1. A detection method based on small-scale pine wood nematode disease trees specifically comprises the following steps:
step 1: acquiring an image, marking the pine wood nematode disease tree, and making into a data set;
step 2: inputting the data set into a bottom layer feature batch processing fusion and multiple feature multiplexing network model, performing feature extraction to obtain an identification model, and identifying the pine wood nematode disease tree;
and step 3: vectorizing the identification result of the pine wood nematode disease tree to obtain longitude and latitude coordinate information of the pine wood nematode disease tree;
and 4, step 4: and uploading the longitude and latitude coordinate information of the pine wood nematode disease tree to a pine wood nematode disease tree supervision platform, checking the geographical distribution condition of the disease tree through the supervision platform, and manually felling.
2. The method according to claim 1, wherein in step 2, the data set is input into a bottom layer feature batch fusion and multiple feature multiplexing network model for feature extraction to obtain a recognition model, and the following steps are specifically adopted:
2-1: selecting and preprocessing an unmanned aerial vehicle image sample set;
2-2: classifying the data sets according to different scales of the disease tree pictures, and separately carrying out migration training on the network model;
2-3: and inputting the data set into a bottom layer feature batch processing fusion and multiple feature multiplexing network model, extracting multiple feature information to obtain a pine wood nematode disease tree identification model, and identifying the pine wood nematode disease tree.
3. The method according to claim 2, wherein in step 2-2, the data sets are classified according to different scales of the disease tree pictures, and the migration training network model is separately performed, and the method comprises the following sub-steps:
2-2-1: in a data set sample with the scale value A, the scale of the disease tree is relatively small, so that the characteristic extraction of the small-scale disease tree by the model is facilitated, the data set with the scale value A is firstly input into a network model, and a pine wood nematode disease tree recognition model is obtained through training;
2-2-2: and then inputting the data set sample with the size of B into a network model, and performing migration training by taking the network model obtained by training with the size of A as a pre-training model for training the data set sample with the size of B to obtain a final pine wood nematode disease tree recognition model.
4. The method according to claim 2, wherein in step 2-3, the building of the underlying feature batch fusion and multiple feature multiplexing network model specifically comprises the following sub-steps:
2-3-1: inputting an image to be detected into a feature extraction backbone network, and extracting features from a shallow layer to a deep layer in a bottom-up mode, wherein the features are as follows:
performing down-sampling operation on the input graph through convolution to obtain a feature graph L1;
performing maximum pooling downsampling operation on the feature map L1 through convolution, and sequentially performing operation on a plurality of convolution residual blocks to obtain a feature map L2; obtaining a feature map L3 by carrying out a plurality of convolution residual block operations on the feature map L2; obtaining a feature map L4 by carrying out a plurality of convolution residual block operations on the feature map L3; obtaining a feature map L5 by carrying out a plurality of convolution residual block operations on the feature map L4;
2-3-2: and (3) carrying out batch processing fusion on the bottom layer features and multiplexing the deep layer features to enhance the feature extraction capability of the model on the pine wood nematode disease trees with different scales, and then identifying the pine wood nematode disease trees.
5. The method according to claim 4, wherein in step 2-3-2, the specific steps are as follows:
performing convolution operation on the bottom layer feature map L1, the feature map L2 and the feature map L3 to enable the channel numbers of the bottom layer feature map L1, the feature map L2 and the feature map L3 to be the same, and then performing feature fusion to obtain a feature map F2;
carrying out convolution operation on the feature map F2 by using convolution, eliminating the mixed superposition effect of the fused features, and obtaining a final predicted feature map Ps1 for detecting the small-scale disease tree;
performing convolution operation on the deep feature maps L4 and L5 to enable the channel numbers of the feature maps L4 and L5 to be the same as the channel number of the feature map L3; then carrying out feature fusion to obtain a feature map F3;
performing convolution operation on the feature map F3 by using convolution, eliminating the mixed superposition effect of the fused features, and obtaining a final predicted feature map Ps2 for detecting a small-scale disease tree which is not detected by the predicted feature map Ps 1;
fusing the feature map L4 with the feature map F2 subjected to the downsampling operation and the feature map L5 subjected to the upsampling operation to obtain a feature map F4;
carrying out convolution operation on the feature map F4 by using convolution, eliminating the mixed superposition effect of the fused features, and obtaining a predicted feature map Pm for detecting a medium-scale disease tree;
and fusing the feature map L5 with the feature map F2 subjected to the downsampling operation to obtain a feature map F5, and performing convolution operation on the feature map F5 to obtain a predicted feature map Pl for detecting the large-scale disease tree.
6. The method according to claim 1, characterized in that step 4 comprises in particular the following sub-steps:
4-1: importing the longitude and latitude coordinate information of the identified diseased tree into a pine wood nematode disease tree supervision platform;
4-2: the geographical distribution condition of diseased trees can be checked on the pine wood nematode diseased tree supervision platform, diseased trees are found according to diseased tree navigation information of the supervision platform, and diseased pines are surveyed and manually felled.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210617043.9A CN115063700A (en) | 2022-06-01 | 2022-06-01 | Detection method based on small-scale pine wood nematode disease tree |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210617043.9A CN115063700A (en) | 2022-06-01 | 2022-06-01 | Detection method based on small-scale pine wood nematode disease tree |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115063700A true CN115063700A (en) | 2022-09-16 |
Family
ID=83198461
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210617043.9A Pending CN115063700A (en) | 2022-06-01 | 2022-06-01 | Detection method based on small-scale pine wood nematode disease tree |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115063700A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116580847A (en) * | 2023-07-14 | 2023-08-11 | 天津医科大学总医院 | Modeling method and system for prognosis prediction of septic shock |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114387528A (en) * | 2021-12-29 | 2022-04-22 | 浙江同创空间技术有限公司 | Pine nematode disease monitoring space-air-ground integrated monitoring method |
| CN114462485A (en) * | 2021-01-29 | 2022-05-10 | 王建新 | Red date jujube witches broom initial-stage control method |
-
2022
- 2022-06-01 CN CN202210617043.9A patent/CN115063700A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114462485A (en) * | 2021-01-29 | 2022-05-10 | 王建新 | Red date jujube witches broom initial-stage control method |
| CN114387528A (en) * | 2021-12-29 | 2022-04-22 | 浙江同创空间技术有限公司 | Pine nematode disease monitoring space-air-ground integrated monitoring method |
Non-Patent Citations (2)
| Title |
|---|
| 刘遐龄;程多祥;李涛;陈小平;高文娟;: "无人机遥感影像的松材线虫病危害木自动监测技术初探", 中国森林病虫, no. 05, 15 September 2018 (2018-09-15) * |
| 徐信罗;陶欢;李存军;程成;郭杭;周静平;: "基于Faster R-CNN的松材线虫病受害木识别与定位", 农业机械学报, no. 07, 23 July 2020 (2020-07-23) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116580847A (en) * | 2023-07-14 | 2023-08-11 | 天津医科大学总医院 | Modeling method and system for prognosis prediction of septic shock |
| CN116580847B (en) * | 2023-07-14 | 2023-11-28 | 天津医科大学总医院 | Method and system for predicting prognosis of septic shock |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107016405B (en) | A kind of pest image classification method based on classification prediction convolutional neural networks | |
| CN112446388A (en) | Multi-category vegetable seedling identification method and system based on lightweight two-stage detection model | |
| CN109800736A (en) | A kind of method for extracting roads based on remote sensing image and deep learning | |
| CN108596248A (en) | A kind of classification of remote-sensing images model based on improvement depth convolutional neural networks | |
| CN107832797B (en) | Multispectral image classification method based on depth fusion residual error network | |
| CN110222767B (en) | Three-dimensional point cloud classification method based on nested neural network and grid map | |
| CN112560716B (en) | High-resolution remote sensing image water body extraction method based on low-level feature fusion | |
| Su et al. | LodgeNet: Improved rice lodging recognition using semantic segmentation of UAV high-resolution remote sensing images | |
| CN103839078A (en) | Hyperspectral image classifying method based on active learning | |
| CN116883853B (en) | Crop space-time information remote sensing classification method based on transfer learning | |
| CN108256557B (en) | Hyperspectral image classification method combining deep learning and neighborhood integration | |
| CN117496345A (en) | CVCUnet-based multi-terrain multi-band farmland extraction method | |
| CN115965819A (en) | Lightweight pest identification method based on Transformer structure | |
| CN115063700A (en) | Detection method based on small-scale pine wood nematode disease tree | |
| Cohen et al. | Development of an automated monitoring platform for invasive plants in a rare Great Lakes ecosystem using uncrewed aerial systems and convolutional neural networks | |
| Zheng et al. | YOLOv4-lite–based urban plantation tree detection and positioning with high-resolution remote sensing imagery | |
| Lin et al. | A novel approach for estimating the flowering rate of litchi based on deep learning and UAV images | |
| CN114241309A (en) | Rice sheath blight identification method and system based on ShuffleNet V2-Unet | |
| CN116543165B (en) | Remote sensing image fruit tree segmentation method based on dual-channel composite depth network | |
| CN117274803A (en) | Tree classification method and system based on DSC-DC convolutional neural network | |
| CN115909066A (en) | Pine wood nematode disease tree detection method based on small target sample expansion and pooling weighting | |
| CN116721385A (en) | Machine learning-based RGB camera data cyanobacteria bloom monitoring method | |
| CN115471478A (en) | Unmanned aerial vehicle remote sensing monitoring method for pine tree diseases based on yolov5 | |
| CN115170987A (en) | Method for detecting diseases of grapes based on image segmentation and registration fusion | |
| CN115063602A (en) | Crop pest and disease identification method based on improved YOLOX-S network |
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 |