CN114972264A - Method and device for identifying mung bean leaf spot based on MS-PLNet model - Google Patents
Method and device for identifying mung bean leaf spot based on MS-PLNet model Download PDFInfo
- Publication number
- CN114972264A CN114972264A CN202210598574.8A CN202210598574A CN114972264A CN 114972264 A CN114972264 A CN 114972264A CN 202210598574 A CN202210598574 A CN 202210598574A CN 114972264 A CN114972264 A CN 114972264A
- Authority
- CN
- China
- Prior art keywords
- convolution
- leaf spot
- plnet
- model
- feature extraction
- 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
- 238000000034 method Methods 0.000 title claims abstract description 32
- 240000004922 Vigna radiata Species 0.000 title claims abstract description 31
- 235000010721 Vigna radiata var radiata Nutrition 0.000 title claims abstract description 31
- 235000011469 Vigna radiata var sublobata Nutrition 0.000 title claims abstract description 31
- 230000004927 fusion Effects 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 14
- 230000008569 process Effects 0.000 claims abstract description 9
- 238000012360 testing method Methods 0.000 claims abstract description 9
- 230000007246 mechanism Effects 0.000 claims abstract description 8
- 238000010276 construction Methods 0.000 claims abstract description 4
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000000605 extraction Methods 0.000 claims description 52
- 201000010099 disease Diseases 0.000 claims description 23
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 23
- 241000196324 Embryophyta Species 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 244000013123 dwarf bean Species 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 238000012795 verification Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000005286 illumination Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 4
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 claims description 2
- 238000003704 image resize Methods 0.000 claims description 2
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 8
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000001514 detection method Methods 0.000 description 4
- 238000005520 cutting process Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000012271 agricultural production Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 241000238631 Hexapoda Species 0.000 description 1
- 241000282414 Homo sapiens Species 0.000 description 1
- 241000607479 Yersinia pestis Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004166 bioassay Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000003306 harvesting Methods 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000009828 non-uniform distribution Methods 0.000 description 1
- 238000000399 optical microscopy Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
- 230000017260 vegetative to reproductive phase transition of meristem Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Multimedia (AREA)
- Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Mining & Mineral Resources (AREA)
- Marketing (AREA)
- Agronomy & Crop Science (AREA)
- Animal Husbandry (AREA)
- Marine Sciences & Fisheries (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Probability & Statistics with Applications (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method and a device for identifying mung bean leaf spot based on an MS-PLNet model, which comprises the following steps: (1) acquiring a mung bean leaf data set for training, verifying and testing, and performing image preprocessing on the data set; (2) constructing an MS-PLNet network model for identifying the mung bean leaf spot, wherein a backbone network in the model is constructed by using common convolution and improved deep separable convolution, and a three-stage strategy of firstly increasing, then reducing and then increasing the number of channels is carried out in the construction process, so that network parameters are reduced; in the feature fusion stage, a channel attention mechanism is used, important channels are concerned according to the importance degree of the channels, and redundant channels are restrained; (3) and training by using the built model to obtain the classifier for detecting the mung bean leaf spot. The method can be used for detecting the mung bean leaf spot and other crop leaf spots, and has the advantages of high recognition rate, good robustness and wide application value.
Description
Technical Field
The invention relates to the field of agricultural plant protection, in particular to a mung bean leaf spot identification method based on an MS-PLNet model.
Background
Crops are various plants cultivated by human beings in agricultural production activities and are divided into two categories, namely grain crops and economic crops, and no matter which crops are easily affected by various aspects such as moisture, soil, plant diseases and insect pests in the planting process, so that the influence of the factors on the crop planting needs to be reduced by using an effective cultivation technology.
Identification of disease is a key step in the agricultural field as it affects crop planting and harvesting. The identification of these diseases is time and resource intensive and often requires expert knowledge. Farmers will have to manually check the symptoms of large crops. This process is repetitive and inefficient. The mung bean leaf spot is used as a main disease of mung beans, the harmful part is a leaf, the damage is serious in the flowering and pod bearing period, water stain-shaped brown spots appear on the leaf at the early stage of disease development, and the green leaf spot is expanded to form a brown-reddish edge and a brown-light middle pale gray-light brown near-round spot. When the disease is serious, the spots fuse into pieces and dry up quickly. The yield of the patient is reduced by 20-50% in the light patients, and the yield is up to 90% in the serious patients. The traditional green bean leaf spot identification has a lot of defects in actual agricultural production, the green bean leaf spot cannot be accurately and objectively evaluated by visual observation, and the green bean leaf spot identification is difficult to realize on a large scale. Optical microscopy, bioassay, pcr, etc. techniques, while capable of accurate identification, require professional operation and are time consuming and labor intensive, and are difficult to apply to real-time online identification of green bean leaf spot. Using computer vision technology, disease identification and classification has improved tremendously, and today, intelligent agriculture is focused on providing infrastructure to take advantage of advanced technologies such as internet of things and big data. With the recent advent of deep learning, these diseases can now be automatically identified with extremely high accuracy.
At present, the convolutional neural network is widely applied to the related field of agricultural engineering, and plays an important role in the aspect of identifying mung bean leaf diseases. In order to achieve better performance, the number of network layers has increased in recent years, from AlexNet at 7 layers to VGGNet at 16 layers, to GoogleNet at 22 layers, to ResNet at 152 layers, to ResNet at thousands of layers, and so on. However, the traditional convolutional neural network identification systems have the problems of large parameters and high computation amount. The purpose of researching the lightweight convolutional neural network is to greatly reduce the calculated amount and parameters of the large convolutional neural network on the basis of ensuring that the accuracy of the large convolutional neural network is equivalent, so as to reduce training and deployment resources.
Disclosure of Invention
The purpose of the invention is as follows: the invention overcomes the defects of large parameter and high model calculation amount of the traditional convolution neural network model, and provides the identifying method and the identifying device for the leaf spot of the mung bean based on the MS-PLNet model, which not only have better leaf spot detection accuracy, but also enable the model to have smaller parameter, and realize the rapid identification and statistical analysis of the leaf spot.
The technical scheme is as follows: the invention provides a mung bean leaf spot identification method based on an MS-PLNet model, which comprises the following steps:
(1) preprocessing plant leaf disease images which are obtained in advance and contain various disease types, and dividing the plant leaf disease images into a training set, a verification set and a test set;
(2) constructing an MS-PLNet leaf spot recognition network;
(3) training an MS-PLNet leaf spot recognition network, comparing an obtained prediction result with a real label through a pre-processed training set through the built MS-PLNet leaf spot recognition network to calculate classification loss, and updating model parameters through a momentum gradient descent algorithm to obtain an MS-PLNet model for leaf spot recognition;
(4) and (4) identifying the leaf spot of the mung bean, inputting the images in the verification set into the MS-PLNet model trained in the step (3) to evaluate the performance of the model, and inputting the images in the test set into the trained model to obtain the leaf spot type of the images through forward propagation.
Further, the image in the step (1) is a mung bean leaf image shot by a common camera, and data enhancement is performed on all leaf images, wherein the data enhancement comprises random illumination transformation, left-right turning, up-down turning, diagonal turning and random cutting.
Further, the MS-PLNet leaf spot identification network comprises a feature extraction module, a feature fusion module and a classifier; inputting the images of three scales into a feature extraction module; constructing feature extraction modules of three scales, wherein the feature extraction modules are constructed by using common convolution and improved depth separable convolution, the number of channels is increased by the first feature extraction module, the number of channels is decreased by the second feature extraction module, and the number of channels is increased by the third feature extraction module in the construction process; inputting the feature graphs obtained by the feature extraction modules of the three scales into a feature fusion module adopting a channel attention mechanism; and inputting the feature graph output by the feature fusion into a classifier to obtain a classification result.
Further, the step (2) comprises the steps of:
(21) the image input stage processes the input 3-channel image resize into 128 x 128 pixels, which is labeled as IM 0; the first feature extraction module inputs a 256 × 256 pixel image after Unsample deconvolution operation, and the image is marked as IM 1; the second feature extraction module inputs a 128 x 128 pixel image, labeled IM 2; the third feature extraction module inputs a 64 × 64 pixel image after MaxPool2d maximum pooling operation, and is marked as IM 3;
(22) in the feature extraction module of the first scale, computing an IM1 as an input of a multi-convolution network with a convolution kernel of 4 × 4 to obtain C11, computing C11 with a convolution kernel of 1 × 1 by adding a channel number and 7 × 7 by deep convolution to obtain C12, computing C12 with the same convolution to obtain C13, computing C13 with the same convolution to obtain C14, computing C14 with a convolution kernel of 1 × 1 by reducing a channel number and 7 × 7 by deep convolution to obtain C15, computing C15 with a convolution kernel of 1 × 1 by adding a channel number and 7 × 7 by deep convolution to obtain C16, and computing C16 with the same convolution to obtain C17; in the feature extraction module of the second scale, IM2 is used as input of a multi-convolution network to perform calculation with convolution kernel of 4 × 4 to obtain C21, C21 is used to perform point convolution with convolution kernel of 1 × 1 to increase the channel number and 5 × 5 to obtain C22, C22 is used to perform the same convolution calculation to obtain C23, C23 is then used to perform the same convolution calculation to obtain C24, C24 is used to perform point convolution with convolution kernel of 1 × 1 to decrease the channel number and 7 × 7 to obtain C25, C25 is used to perform point convolution with convolution kernel of 1 × 1 to increase the channel number and 7 × 7 to obtain C26, and C26 is used to perform the same convolution calculation to obtain C27; in a third scale feature extraction module, calculating an IM3 as an input of a multi-convolution network with a convolution kernel of 4 × 4 to obtain C31, calculating C31 with a convolution kernel of 1 × 1 by adding a channel number and 3 × 3 by deep convolution to obtain C32, calculating C32 with the same convolution to obtain C33, calculating C33 with the same convolution to obtain C34, calculating C34 with a convolution kernel of 1 × 1 by reducing a channel number and 3 × 3 by deep convolution to obtain C35, calculating C35 with a convolution kernel of 1 × 1 by adding a channel number and 3 × 3 by deep convolution to obtain C36, and calculating C36 with the same convolution to obtain C37;
(23) performing feature fusion on feature maps C17, C27 and C37 obtained by the three-scale feature extraction module through a concatenate operation to obtain R1, readjusting the weight of each feature channel of the R1 through a channel attention mechanism, and reasonably recombining high-dimensional features to obtain R2;
(24) and inputting R2 into the average pooling layer, the full-link layer and the Softmax activation function, and outputting a classification result.
Based on the same inventive concept, the invention also provides a device for identifying the mung bean leaf spot based on the MS-PLNet model, which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the computer program realizes the identification method for the mung bean leaf spot based on the MS-PLNet model when being loaded into the processor.
Has the advantages that: compared with the prior art, the invention has the beneficial effects that: the method constructs an MS-PLNet model facing the green bean leaf spot identification, and in a data input stage, multi-scale images are input to increase the diversity of model extraction features; in the feature extraction stage, a common convolution and an improved depth separable convolution are used for building, and a three-stage strategy of increasing the number of channels, then reducing the number of channels and finally increasing the number of channels is adopted to reduce network parameters; finally, continuously using a channel attention mechanism in the characteristic fusion stage, paying attention to an important channel according to the importance degree of the channel, and inhibiting a redundant channel; the identification rate of the invention is 99.96%, the size of the model is 7.59M, the invention can be used for detecting the mung bean leaf spot and other crop leaf spots, and has better robustness and wide application value.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of the MS-PLNet model structure;
FIG. 3 is a schematic diagram of a feature extraction network structure of scale one in the MS-PLNet model;
FIG. 4 is a schematic diagram of a feature extraction network structure of scale two in the MS-PLNet model;
FIG. 5 is a schematic diagram of a feature extraction network structure of scale three in the MS-PLNet model;
FIG. 6 is a graph comparing the performance of an experiment using the present invention with a lightweight model.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
A large number of variables are involved in the present embodiment, and each variable will now be described as shown in table 1.
Table 1 description of variables
The invention provides a mung bean leaf spot identification method based on an MS-PLNet model, which specifically comprises the following steps of:
step 1: the method comprises the steps of preprocessing plant leaf disease images which are obtained in advance and contain various disease types, and dividing the images into a training set, a verification set and a test set.
Shooting the mung bean leaves by using a Canon EOS 550D camera, and performing data enhancement on all leaf images, wherein the data enhancement comprises random illumination transformation, left-right turning, up-down turning, diagonal turning and random cutting.
The data set used by the invention is a PlantVillage public data set and a mung bean leaf data set shot by Jiangsu province academy of agricultural sciences under a natural scene for experiments. Wherein the plantarvillege public dataset comprised a total of 20636 images of 15 different classes of diseased and healthy leaf blades; the data set provided by the farm institute of Jiangsu province included 1006 images of diseased and healthy mung bean leaves. In order to make up the influence of non-uniform distribution of the plant Village data samples on the model identification performance and avoid network overfitting, the method carries out data enhancement operation before training. The data enhancement methods used are: 1) random illumination transformation: PCA (principal component analysis) is carried out on pixels of all images so as to obtain characteristic values and characteristic vectors in an RGB space, and random addition and subtraction are carried out on the pixel values of the images along the direction of the characteristic vectors so as to simulate images with different illuminations. 2) And (4) random overturning: including left-right flipping, up-down flipping, and diagonal flipping. 3) Random cutting: the data enhancement mode can move the position of each area on the picture on the basis of reserving the image proportion. Finally unifying the size of all images into 128 × 128 pixels. After pretreatment, a total of 31117 final image samples were obtained as the mung bean leaf spot dataset of the final experiment.
Step 2: constructing an MS-PLNet leaf spot recognition network, as shown in FIG. 2, comprising a feature extraction module, a feature fusion module and a classifier; inputting the images of three scales into a feature extraction module; constructing feature extraction modules of three scales, wherein the feature extraction modules are constructed by using common convolution and improved depth separable convolution, the number of channels is increased by the first feature extraction module, the number of channels is decreased by the second feature extraction module, and the number of channels is increased by the third feature extraction module in the construction process; and inputting the feature maps obtained by the feature extraction modules of the three scales into a feature fusion module adopting a channel attention mechanism. And on the basis of the obtained feature fusion, classifying and identifying the disease image through a classifier consisting of an average pooling layer and a linear layer. The method can realize the extraction of the image characteristics from three different angles in a multi-scale characteristic extraction mode on the basis of not increasing model parameters, and is suitable for the classification of the leaf spot of the crops. The method specifically comprises the following steps:
(1) the multi-scale feature extraction network is formed by adding an ordinary 4 multiplied by 4 convolution and an improved depth separable convolution, and correspondingly reduces the number of channels on the premise of not losing a feature map of an effective channel through a three-stage strategy, namely, the number of channels is increased firstly and then is reduced, and the multi-scale feature extraction network process is as follows:
1) the feature extraction network of scale one is shown in fig. 3: an 18-layer multi-combination network is constructed, and 16 convolution layers and 2 deconvolution layers are formed in total; the input image is a color image of 3 channels, and the resolution is 128 multiplied by 128; the convolution with convolution kernel of 1 × 1 is used to maintain the number of channels to be 3; the image resolution is increased to 256 × 256 using a deconvolution operation with a convolution kernel of 2 × 2; reducing the resolution of the feature map to 64 × 64 and the number of channels to 96 by using convolution with a convolution kernel of 4 × 4; using convolution with a convolution kernel of 1 × 1, the number of channels increases to 192; convolution with convolution kernel 7 × 7 and groups 192 reduces the image resolution to 32 × 32; using convolution with a convolution kernel of 1 × 1, the number of channels increases to 384; the image resolution is reduced to 16 × 16 using convolution with convolution kernels of 7 × 7 and groups of 384; convolution with convolution kernel of 1 × 1 is used, and the number of channels is increased to 768; the image resolution is reduced to 8 × 8 by using convolution with convolution kernels of 7 × 7 and groups of 768; using convolution with a convolution kernel of 1 × 1, the number of channels is reduced to 384; convolution with convolution kernel 7 × 7 and groups 384 reduces the image resolution to 4 × 4; convolution with convolution kernel of 1 × 1 is used, and the number of channels is increased to 768; convolution with convolution kernel of 7 × 7 and groups of 768 is used to reduce the image resolution to 2 × 2; convolution with convolution kernel of 1 × 1 is used, and the number of channels is maintained at 768; the image resolution is increased to 4 × 4 using a deconvolution operation with a convolution kernel of 2 × 2; convolution with convolution kernel of 7 × 7 and groups of 768 is used to reduce the image resolution to 2 × 2; convolution with convolution kernel 7 × 7 and groups 768 reduces the image resolution to 1 × 1.
2) The feature extraction network of scale two is shown in fig. 4: a 17-layer multi-combination network is constructed, wherein 15 convolution layers and 2 deconvolution layers are formed; the input image is a color image of 3 channels, and the resolution is 128 multiplied by 128; reducing the resolution of the characteristic diagram to be 32 multiplied by 32 and the number of channels to be 96 by using convolution with convolution kernel of 4 multiplied by 4; using convolution with a convolution kernel of 1 × 1, the number of channels increases to 192; the image resolution is reduced to 16 × 16 using convolution with convolution kernel of 5 × 5 and groups of 192; using convolution with a convolution kernel of 1 × 1, the number of channels increases to 384; the image resolution is reduced to 8 × 8 using convolution with a convolution kernel of 5 × 5 and groups of 384; convolution with convolution kernel of 1 × 1 is used, and the number of channels is increased to 768; convolution with convolution kernel of 5 × 5 and groups of 768 is used to reduce the image resolution to 4 × 4; using convolution with a convolution kernel of 1 × 1, the number of channels is reduced to 384; the image resolution is reduced to 2 × 2 using convolution with a convolution kernel of 5 × 5 and groups of 384; convolution with convolution kernel of 1 × 1 is used, and the number of channels is increased to 768; the image resolution is reduced to 1 × 1 using convolution with convolution kernel of 5 × 5 and groups of 768; convolution with convolution kernel of 1 × 1 is used, and the number of channels is maintained at 768; using deconvolution with a convolution kernel of 2 x 2, the image resolution is increased to 2 x 2; the image resolution is reduced to 1 × 1 using convolution with convolution kernel of 5 × 5 and groups of 768; convolution with convolution kernel of 1 × 1 is used, and the number of channels is maintained at 768; using deconvolution with a convolution kernel of 2 x 2, the image resolution is increased to 2 x 2; convolution with convolution kernel of 5 × 5 and groups of 768 reduces the image resolution to 1 × 1.
3) The feature extraction network of scale three is shown in fig. 5: the method comprises the following steps of constructing a multi-combination network with 20 layers, wherein 16 convolutional layers, 3 anti-convolutional layers and 1 maximum pooling layer are formed in total; the input image is a color image of 3 channels, and the resolution is 128 multiplied by 128; maintaining the number of channels to be 3 by convolution with a convolution kernel of 1 × 1; reducing the image resolution to 64 x 64 using a maximum pooling operation with a convolution kernel of 2 x 2; convolution with convolution kernel of 4 × 4 is used to reduce the resolution of the feature map to 16 × 16 and the number of channels to 96; using convolution with a convolution kernel of 1 × 1, the number of channels increases to 192; the image resolution is reduced to 8 × 8 using convolution with convolution kernel of 3 × 3 and groups of 192; using convolution with a convolution kernel of 1 × 1, the number of channels increases to 384; the image resolution is reduced to 4 × 4 using convolution with convolution kernel of 3 × 3 and groups of 384; convolution with convolution kernel of 1 × 1 is used, and the number of channels is increased to 768; the image resolution is reduced to 2 x 2 by using convolution with a convolution kernel of 3 x 3 and groups of 768; using convolution with a convolution kernel of 1 × 1, the number of channels is reduced to 384; the image resolution is reduced to 1 × 1 using convolution with a convolution kernel of 3 × 3 and groups of 384; convolution with convolution kernel of 1 × 1 is used, and the number of channels is increased to 768; deconvolution with a convolution kernel of 2 x 2 was used to increase the image resolution to 2 x 2; the image resolution is reduced to 1 × 1 by using convolution with a convolution kernel of 3 × 3 and groups of 768; convolution with convolution kernel of 1 × 1 is used, and the number of channels is maintained at 768; deconvolution with a convolution kernel of 2 x 2 was used to increase the image resolution to 2 x 2; the image resolution is reduced to 1 × 1 by using convolution with a convolution kernel of 3 × 3 and groups of 768; convolution with convolution kernel of 1 × 1 is used, and the number of channels is maintained at 768; deconvolution with a convolution kernel of 2 x 2 was used to increase the image resolution to 2 x 2; convolution with convolution kernel 3 × 3 and groups 768 reduces the image resolution to 1 × 1.
(2) And the feature fusion network performs feature fusion on the feature graph obtained by the feature extraction network through a channel attention mechanism. And accumulating the characteristic diagrams obtained from the three characteristic extraction modules in the step one to obtain 2304 characteristic diagrams, and inputting the characteristic diagrams into the SqEx module to obtain a concerned channel.
(3) The classification module is further input into an average pooling layer with a convolution kernel of 1 multiplied by 1; 2304 vectors are input into the Linear layer, the output of which is the number of classes of samples.
And step 3: training an MS-PLNet leaf spot recognition network, comparing the obtained prediction result with a real label through a built MS-PLNet leaf spot recognition network by using a preprocessed training set, calculating classification loss, updating model parameters through a momentum gradient descent algorithm, and obtaining an MS-PLNet model for leaf spot recognition.
And 4, step 4: and (3) identifying the leaf spot of the mung bean, inputting the images in the verification set into the MS-PLNet model trained in the step (3) to evaluate the performance of the model, and inputting the images in the test set into the trained model to obtain the leaf disease types of the images through forward propagation.
Based on the same inventive concept, the invention also provides a device for identifying the mung bean leaf spot based on the MS-PLNet model, which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the computer program realizes the identification method for the mung bean leaf spot based on the MS-PLNet model when being loaded into the processor.
The experimental result of the MS-PLNet model is shown in FIG. 6, the embodiment verifies on the verification set, tests on the test set, and it can be seen that the algorithm of the patent can bring about 99.96% of the detection accuracy rate of the leaf spot, which is close to 100%, and meanwhile, the algorithm of the patent has better generalization and faster convergence rate during training.
The MS-PLNet model for identifying the green bean leaf spot disease is compared with the performance of the existing popular lightweight algorithm, the detection accuracy of the leaf spot disease of 99.96% is achieved on the premise that the model parameter is only 7.59M, the detection accuracy is far higher than that of the existing popular lightweight algorithm, and the crop leaf spot disease can be better detected.
Claims (5)
1. A mung bean leaf spot identification method based on an MS-PLNet model is characterized by comprising the following steps:
(1) preprocessing plant leaf disease images which are obtained in advance and contain various disease types, and dividing the plant leaf disease images into a training set, a verification set and a test set;
(2) constructing an MS-PLNet leaf spot recognition network;
(3) training an MS-PLNet leaf spot recognition network, comparing an obtained prediction result with a real label through a pre-processed training set through the built MS-PLNet leaf spot recognition network to calculate classification loss, and updating model parameters through a momentum gradient descent algorithm to obtain an MS-PLNet model for leaf spot recognition;
(4) and (4) identifying the leaf spot of the mung bean, inputting the images in the verification set into the MS-PLNet model trained in the step (3) to evaluate the performance of the model, and inputting the images in the test set into the trained model to obtain the leaf spot type of the images through forward propagation.
2. The identification method for the mung bean leaf spot disease based on the MS-PLNet model according to claim 1, wherein the image in the step (1) is a mung bean leaf image shot by a common camera, and data enhancement is performed on all the leaf images, wherein the data enhancement comprises random illumination transformation, left-right turning, up-down turning, diagonal turning and random clipping.
3. The MS-PLNet model-based green bean leaf spot recognition method according to claim 1, wherein the MS-PLNet leaf spot recognition network comprises a feature extraction module, a feature fusion module and a classifier; inputting the images of three scales into a feature extraction module; constructing feature extraction modules of three scales, wherein the feature extraction modules are constructed by using common convolution and improved depth separable convolution, the number of channels is increased by the first feature extraction module, the number of channels is decreased by the second feature extraction module, and the number of channels is increased by the third feature extraction module in the construction process; inputting the feature graphs obtained by the feature extraction modules of the three scales into a feature fusion module adopting a channel attention mechanism; and inputting the feature graph output by the feature fusion into a classifier to obtain a classification result.
4. The MS-PLNet model-based green bean leaf spot identification method according to claim 1, wherein the step (2) comprises the steps of:
(21) the image input stage processes the input 3-channel image resize into 128 × 128 pixels, which is labeled as IM 0; the first feature extraction module inputs a 256 × 256 pixel image after Unsample deconvolution operation, and the image is marked as IM 1; the second feature extraction module inputs a 128 x 128 pixel image, labeled IM 2; the third feature extraction module inputs a 64 × 64 pixel image after MaxPool2d maximum pooling operation, and is marked as IM 3;
(22) in the feature extraction module of the first scale, computing an IM1 as an input of a multi-convolution network with a convolution kernel of 4 × 4 to obtain C11, computing C11 with a convolution kernel of 1 × 1 to increase the number of channels and performing depth convolution with a convolution kernel of 7 × 7 to obtain C12, computing C12 with the same convolution to obtain C13, computing C13 with the same convolution to obtain C14, computing C14 with a convolution kernel of 1 × 1 to decrease the number of channels and performing depth convolution with a convolution kernel of 7 × 7 to obtain C15, computing C15 with a convolution kernel of 1 × 1 to increase the number of channels and performing depth convolution with a convolution kernel of 7 × 7 to obtain C16, and computing C16 with the same convolution to obtain C17; in the feature extraction module of the second scale, IM2 is used as input of a multi-convolution network to perform calculation with convolution kernel of 4 × 4 to obtain C21, C21 is used to perform point convolution with convolution kernel of 1 × 1 to increase the channel number and 5 × 5 to obtain C22, C22 is used to perform the same convolution calculation to obtain C23, C23 is then used to perform the same convolution calculation to obtain C24, C24 is used to perform point convolution with convolution kernel of 1 × 1 to decrease the channel number and 7 × 7 to obtain C25, C25 is used to perform point convolution with convolution kernel of 1 × 1 to increase the channel number and 7 × 7 to obtain C26, and C26 is used to perform the same convolution calculation to obtain C27; in a third scale feature extraction module, calculating an IM3 as an input of a multi-convolution network with a convolution kernel of 4 × 4 to obtain C31, calculating C31 with a convolution kernel of 1 × 1 by adding a channel number and 3 × 3 by deep convolution to obtain C32, calculating C32 with the same convolution to obtain C33, calculating C33 with the same convolution to obtain C34, calculating C34 with a convolution kernel of 1 × 1 by reducing a channel number and 3 × 3 by deep convolution to obtain C35, calculating C35 with a convolution kernel of 1 × 1 by adding a channel number and 3 × 3 by deep convolution to obtain C36, and calculating C36 with the same convolution to obtain C37;
(23) performing feature fusion on feature maps C17, C27 and C37 obtained by three-scale feature extraction modules through a concatenate operation to obtain R1, readjusting the weight of each feature channel of the R1 through a channel attention mechanism, and reasonably recombining high-dimensionality features to obtain R2;
(24) and inputting R2 into the average pooling layer, the full-link layer and the Softmax activation function, and outputting a classification result.
5. An apparatus for identifying a mung bean leaf spot based on an MS-PLNet model, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the computer program when loaded into the processor implements the method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210598574.8A CN114972264A (en) | 2022-05-30 | 2022-05-30 | Method and device for identifying mung bean leaf spot based on MS-PLNet model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210598574.8A CN114972264A (en) | 2022-05-30 | 2022-05-30 | Method and device for identifying mung bean leaf spot based on MS-PLNet model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114972264A true CN114972264A (en) | 2022-08-30 |
Family
ID=82957358
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210598574.8A Pending CN114972264A (en) | 2022-05-30 | 2022-05-30 | Method and device for identifying mung bean leaf spot based on MS-PLNet model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114972264A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116824276A (en) * | 2023-08-29 | 2023-09-29 | 广东海洋大学 | Method for identifying ocean vortex edge chlorophyll ring based on convolutional neural network |
-
2022
- 2022-05-30 CN CN202210598574.8A patent/CN114972264A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116824276A (en) * | 2023-08-29 | 2023-09-29 | 广东海洋大学 | Method for identifying ocean vortex edge chlorophyll ring based on convolutional neural network |
| CN116824276B (en) * | 2023-08-29 | 2023-12-12 | 广东海洋大学 | Method for identifying ocean vortex edge chlorophyll ring based on convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Naikwadi et al. | Advances in image processing for detection of plant diseases | |
| da Silva et al. | Estimating soybean leaf defoliation using convolutional neural networks and synthetic images | |
| CN114766041A (en) | System and method for determining crop damage | |
| CN113657294B (en) | Crop disease and insect pest detection method and system based on computer vision | |
| Ninomiya | High-throughput field crop phenotyping: current status and challenges | |
| Hao et al. | Growing period classification of Gynura bicolor DC using GL-CNN | |
| CN113657326A (en) | Weed detection method based on multi-scale fusion module and feature enhancement | |
| CN115272828A (en) | Intensive target detection model training method based on attention mechanism | |
| CN114972208A (en) | YOLOv 4-based lightweight wheat scab detection method | |
| Zhang et al. | Robust image segmentation method for cotton leaf under natural conditions based on immune algorithm and PCNN algorithm | |
| Kumar et al. | An identification of crop disease using image segmentation | |
| CN116030348A (en) | LS-YOLOv5 network-based mung bean leaf spot disease detection method and device | |
| CN114972264A (en) | Method and device for identifying mung bean leaf spot based on MS-PLNet model | |
| Devisurya et al. | Early detection of major diseases in turmeric plant using improved deep learning algorithm | |
| Hossain et al. | A convolutional neural network approach to recognize the insect: A perspective in Bangladesh | |
| Rakesh et al. | Explainable AI for Crop disease detection | |
| Brar et al. | A smart approach to coconut leaf spot disease classification using computer vision and deep learning technique | |
| Liu et al. | “Is this blueberry ripe?”: a blueberry ripeness detection algorithm for use on picking robots | |
| Rony et al. | BottleNet18: Deep Learning-Based Bottle Gourd Leaf Disease Classification | |
| Gao et al. | Classification Method of Rape Root Swelling Disease Based on Convolution Neural Network | |
| Chang et al. | Improved deep learning-based approach for real-time plant species recognition on the farm | |
| Deng et al. | A paddy field segmentation method combining attention mechanism and adaptive feature fusion | |
| Dahiya et al. | An Effective Detection of Litchi Disease using Deep Learning | |
| CN111986149A (en) | Plant disease and insect pest detection method based on convolutional neural network | |
| CN117173122B (en) | Lightweight ViT-based image leaf density determination method and device |
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 |