CN112749667A - Deep learning-based nematode classification and identification method - Google Patents

Deep learning-based nematode classification and identification method Download PDF

Info

Publication number
CN112749667A
CN112749667A CN202110060316.XA CN202110060316A CN112749667A CN 112749667 A CN112749667 A CN 112749667A CN 202110060316 A CN202110060316 A CN 202110060316A CN 112749667 A CN112749667 A CN 112749667A
Authority
CN
China
Prior art keywords
nematode
image
deep learning
learning network
classification
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.)
Granted
Application number
CN202110060316.XA
Other languages
Chinese (zh)
Other versions
CN112749667B (en
Inventor
庄佳衍
劳思思
朱莹
肖江剑
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ningbo Institute of Material Technology and Engineering of CAS
Original Assignee
Ningbo Institute of Material Technology and Engineering of CAS
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Ningbo Institute of Material Technology and Engineering of CAS filed Critical Ningbo Institute of Material Technology and Engineering of CAS
Priority to CN202110060316.XA priority Critical patent/CN112749667B/en
Publication of CN112749667A publication Critical patent/CN112749667A/en
Application granted granted Critical
Publication of CN112749667B publication Critical patent/CN112749667B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
    • G06V20/695—Preprocessing, e.g. image segmentation
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
    • G06V20/698—Matching; Classification

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Multimedia (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a nematode classification and identification method based on deep learning, which is characterized in that an attention mask image corresponding to each image is made for a nematode image training set consisting of images of pine wood nematodes and pseudopine wood nematodes, a new attention loss mechanism is added into a deep learning network, the learning and training of an effective area at the tail of the nematode are emphasized, a cross entropy loss function obtained by the images through a network full-connection layer is combined with the new attention loss mechanism and then fed back to the deep learning network, the nematode image set is trained, a nematode image classification and identification reinforced model based on the deep learning network is established, and the effective classification and identification of the pine wood nematodes and the pseudopine wood nematodes are realized.

Description

Deep learning-based nematode classification and identification method
Technical Field
The invention relates to the field of classification detection of deep learning images, in particular to a nematode classification and identification method.
Background
In recent years, with the rapid development of deep learning technology, a feature extraction algorithm for target detection gradually evolves from a traditional algorithm to a detection algorithm based on a neural network, and with the acceleration of modern information construction, how to screen out an effective part from mass data information generated in various fields such as biology, traffic, medical treatment and the like for feature learning is a research focus of the industry at present. For rare data such as images of the pine wood nematodes, the effective part of the image is mainly concentrated on the tail of the pine wood nematodes, and the classification accuracy of the image is poor due to the attention mechanism of the existing deep learning neural network.
Disclosure of Invention
In order to solve the defects of the prior art, the invention provides a nematode classification and identification method based on deep learning.
In order to achieve the purpose, the invention adopts the following technical scheme:
a deep learning-based nematode classification and identification method is characterized by comprising the following steps:
1) establishing two nematode image sets of the pine wood nematodes and the pine wood nematodes, and selecting partial images from the two nematode image sets to form a nematode image training set;
2) making a corresponding attention mask image for each image in the nematode image training set, wherein the attention mask image enhances the attention to the tail of the nematode;
3) extracting the characteristic vector of each image in the nematode image training set by using a deep learning network, carrying out pixel-by-pixel calculation on the characteristic vector of each image and the attention mask image corresponding to the image, and calculating the resultDifferencing with the feature vector of the image to obtain a loss function Lnew；
4) Will lose function LnewAdding the images into the deep learning network, and learning and training the images in the nematode image training set;
5) by adding a loss function LnewThe deep learning network carries out nematode image category prediction on the nematode images which are learned and trained in the step 4), and the predicted nematode image category probability value and nematode image category true value are calculated to obtain a loss function Lclassify；
6) Will lose function LclassifyAdding to said added loss function LnewIn the deep learning network, the nematode image training set is used for training the deep learning network to obtain a nematode classification recognition strengthening model based on deep learning;
7) and classifying and identifying the images of the pine wood nematodes and the pseudopine wood nematodes by using the nematode classification and identification reinforced model based on deep learning.
Preferably, the method for manufacturing the attention mask pattern includes: and (3) setting the pixels of the tail part image containing obvious features in the nematode image as 1 and the pixels of the other body part images as 0, and making an attention mask image corresponding to the nematode image.
Preferably, the deep learning network is a VGG16 deep learning network.
Preferably, the loss function LnewAs follows:
Figure BDA0002900155080000021
wherein the content of the first and second substances,
Figure BDA0002900155080000022
the feature vectors of the image are extracted through a deep learning network, and the Attention Mask is an Attention Mask image corresponding to the image.
Preferably, the loss function LclassifyAs follows:
Lclassify＝-ylabel log yc-(1-ylabel)log(1-yc)
wherein, ylabelIs the true value of nematode image class, ycIs the predicted nematode image class probability value.
The invention discloses a nematode classification and identification method based on deep learning, which aims at producing an attention mask image corresponding to each image in a nematode image training set consisting of images of pine wood nematodes and pseudopine wood nematodes, emphasizes the learning and training of effective regions at tail parts of nematodes by adding a new attention loss mechanism in a deep learning network, combines a cross entropy loss function obtained by the images through a network full-connection layer with the new attention loss mechanism, feeds back the cross entropy loss function to the deep learning network, trains nematode image sets, establishes a nematode image classification and identification reinforced model based on the deep learning network, and realizes the effective classification and identification of the pine wood nematodes and the pseudopine wood nematodes.
Compared with the prior art, the technical scheme of the invention has the following advantages:
1. the large number of image labeling processes are complicated and trivial, and errors are easy to occur.
2. By introducing a new attention mechanism into the deep learning network and enhancing the learning and training of the deep learning network on the tail area of the nematode, the accuracy of the model is improved, and the classification accuracy is improved.
3. The novel attention mechanism strengthens the learning and training of the deep learning network on the image effective area characteristics, can be utilized in similar model training and has portability.
Drawings
FIG. 1 is a schematic flow chart of the deep learning-based nematode classification and identification method of the present invention;
FIG. 2 is an attention mask for the tail region of Bursaphelenchus xylophilus.
Detailed Description
The following detailed description of the preferred embodiments of the present invention, taken in conjunction with the accompanying drawings, will make the advantages and features of the invention easier to understand by those skilled in the art, and thus will clearly and clearly define the scope of the invention.
A deep learning-based nematode classification and identification method is characterized by comprising the following steps:
step (1): and establishing an image set required by nematode classification and identification.
Collecting a large number of nematode pictures, classifying the nematodes according to the species of the nematodes, and selecting partial pictures (clear in shooting) from the two kinds of pictures to form a nematode image training set.
In this embodiment, 2000 images of two different types of nematodes are collected, each of the images of each type of nematodes includes 1000 images, and 500 images are respectively selected from each type of nematode images to form a nematode image training set.
Step (2): an attention mask pattern is made.
Aiming at the nematode image training set obtained in the step 1), in order to remove the interference of other body parts of the nematode and enable the model to only focus on the tail part which is an effective part with obvious characteristics, the pixels of the other body parts of the nematode image are set as 0, the pixels of the tail part image with the obvious characteristics are set as 1, and an Attention Mask image Attention Mask corresponding to each image is manufactured.
And (3): calculating LnewA loss function.
Extracting the characteristic vector of each picture in the nematode image training set, namely the predicted output value by utilizing five convolutional network layers and pooling layers in the VGG16 deep learning network structure
Figure BDA0002900155080000031
To pair
Figure BDA0002900155080000032
And the Attention Mask performs pixel-by-pixel multiplication calculation, and the calculation result is then compared with the predicted output value
Figure BDA0002900155080000033
Differencing to obtain a loss function Lnew：
Figure BDA0002900155080000034
And (4): a new attention mechanism is added for the deep learning network.
The loss function L obtained in the step 3) is usednewFeeding back to the convolution layer of the VGG16 deep learning network by adding a loss function LnewThe reinforcing network learns and trains the nematode images in the training set, and a loss function LnewThe addition of (2) enhances learning and training of the tail region of the nematode.
And (5): predictive classification based on neural networks.
Adding the image learned and trained in the step 4) into a full connection layer of a VGG16 deep learning network, predicting nematode image categories to obtain nematode image category probability value ycThe predicted nematode image category probability value y is obtainedcCalculating with nematode image category true value (nematode image category true value is known parameter) to obtain loss function Lclassify：
Lclassify＝-ylabellog yc-(1-ylabel)log(1-yc)
Wherein, ylabelIs the true value of nematode image class, ycIs the predicted nematode image class probability value. The calculated values are then transmitted back to the convolutional layer through the fully connected layer for learning and training.
And (6): and training the optimized VGG16 deep learning network by using the nematode image training set to obtain a nematode classification and identification reinforced model based on deep learning.
And (7): and classifying and identifying the images of the pine wood nematodes and the pseudopine wood nematodes by using the nematode classification identification strengthening model based on deep learning.
The essence of this embodiment is to put the total loss function L including the new attention loss mechanism into Lnew+LclassifyA deep learning network is introduced, and a nematode classification recognition strengthening model based on deep learning is obtained by training the network, so that the recognition capability of the network on the tail region of the nematode is improved, the pine wood nematode and the pine wood nematode can be accurately recognized, and the accuracy of classification recognition of the pine wood nematode and the pine wood nematode is improved.
The deep learning network model adopted in this embodiment is VGG16, and the finally obtained deep learning classification network is composed of a plurality of convolutional layers and fully connected layers.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes performed by the present specification and drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (7)

1. A deep learning-based nematode classification and identification method is characterized by comprising the following steps:
1) establishing two nematode image sets of the pine wood nematodes and the pine wood nematodes, and selecting partial images from the two nematode image sets to form a nematode image training set;
2) making a corresponding attention mask image for each image in the nematode image training set, wherein the attention mask image enhances the attention to the tail of the nematode;
3) extracting the characteristic vector of each image in the nematode image training set by using a deep learning network, carrying out pixel-by-pixel calculation on the characteristic vector of each image and the attention mask image corresponding to the image, and carrying out difference on the calculation result and the characteristic vector of the image to obtain a loss function Lnew；
4) Will lose function LnewAdding the images into the deep learning network, and learning and training the images in the nematode image training set;
5) by adding a loss function LnewThe deep learning network carries out nematode image category prediction on the nematode images which are learned and trained in the step 4), and the nematode image category probability value and the nematodes which are obtained by predictionCalculating the true value of the image category to obtain a loss function Lclassify；
6) Will lose function LclassifyAdding to said added loss function LnewIn the deep learning network, the nematode image training set is used for training the deep learning network to obtain a nematode classification recognition strengthening model based on deep learning;
7) and classifying and identifying the images of the pine wood nematodes and the pseudopine wood nematodes by using the nematode classification and identification reinforced model based on deep learning.
2. The deep learning-based nematode classification and identification method according to claim 1, wherein said method for making an attention mask map comprises: and (3) setting the pixels of the tail part image containing obvious features in the nematode image as 1 and the pixels of the other body part images as 0, and making an attention mask image corresponding to the nematode image.
3. The deep learning based nematode classification and identification method of claim 1 wherein said deep learning network is a VGG16 deep learning network.
4. The deep learning-based nematode classification and identification method of claim 1 wherein said loss function L isnewThe method added to the deep learning network is as follows: will lose function LnewAnd feeding back to the convolution layer of the deep learning network.
5. The deep learning-based nematode classification and identification method of claim 1 wherein said loss function L isclassifyAdding to said added loss function LnewThe method in the deep learning network comprises the following steps: will lose function LclassifyFeeding back to the convolution layer of the deep learning network and the loss function L through the full connection layer of the deep learning networknewThe sum constitutes the total loss function of the deep learning network.
6. The deep learning-based nematode classification and identification method of claim 1 wherein said loss function LnewAs follows:
Figure FDA0002900155070000021
wherein the content of the first and second substances,
Figure FDA0002900155070000022
the feature vectors of the image are extracted through a deep learning network, and the Attention Mask is an Attention Mask image corresponding to the image.
7. The deep learning-based nematode classification and identification method of claim 1 wherein said loss function LclassifyAs follows:
Lclassify＝-ylabellogyc-(1-ylabel)log(1-yc)
wherein, ylabelIs the true value of nematode image class, ycIs the predicted nematode image class probability value.
CN202110060316.XA 2021-01-15 2021-01-15 Deep learning-based nematode classification and identification method Active CN112749667B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110060316.XA CN112749667B (en) 2021-01-15 2021-01-15 Deep learning-based nematode classification and identification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110060316.XA CN112749667B (en) 2021-01-15 2021-01-15 Deep learning-based nematode classification and identification method

Publications (2)

Publication Number Publication Date
CN112749667A true CN112749667A (en) 2021-05-04
CN112749667B CN112749667B (en) 2023-04-07

Family

ID=75652270

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110060316.XA Active CN112749667B (en) 2021-01-15 2021-01-15 Deep learning-based nematode classification and identification method

Country Status (1)

Country Link
CN (1) CN112749667B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113256636A (en) * 2021-07-15 2021-08-13 北京小蝇科技有限责任公司 Bottom-up parasite species development stage and image pixel classification method

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109255044A (en) * 2018-08-31 2019-01-22 江苏大学 A kind of image intelligent mask method based on YOLOv3 deep learning network
CN110502987A (en) * 2019-07-12 2019-11-26 山东农业大学 A kind of plant pest recognition methods and system based on deep learning
WO2019240900A1 (en) * 2018-06-12 2019-12-19 Siemens Aktiengesellschaft Attention loss based deep neural network training
CN111027436A (en) * 2019-12-03 2020-04-17 吉林大学 Northeast black fungus disease and pest image recognition system based on deep learning
CN111259850A (en) * 2020-01-23 2020-06-09 同济大学 Pedestrian re-identification method integrating random batch mask and multi-scale representation learning
CN111507334A (en) * 2019-01-30 2020-08-07 中国科学院宁波材料技术与工程研究所 Example segmentation method based on key points
WO2020219757A1 (en) * 2019-04-23 2020-10-29 The Johns Hopkins University Abdominal multi-organ segmentation with organ-attention networks
CN112130200A (en) * 2020-09-23 2020-12-25 电子科技大学 Fault identification method based on grad-CAM attention guidance

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019240900A1 (en) * 2018-06-12 2019-12-19 Siemens Aktiengesellschaft Attention loss based deep neural network training
CN109255044A (en) * 2018-08-31 2019-01-22 江苏大学 A kind of image intelligent mask method based on YOLOv3 deep learning network
CN111507334A (en) * 2019-01-30 2020-08-07 中国科学院宁波材料技术与工程研究所 Example segmentation method based on key points
WO2020219757A1 (en) * 2019-04-23 2020-10-29 The Johns Hopkins University Abdominal multi-organ segmentation with organ-attention networks
CN110502987A (en) * 2019-07-12 2019-11-26 山东农业大学 A kind of plant pest recognition methods and system based on deep learning
CN111027436A (en) * 2019-12-03 2020-04-17 吉林大学 Northeast black fungus disease and pest image recognition system based on deep learning
CN111259850A (en) * 2020-01-23 2020-06-09 同济大学 Pedestrian re-identification method integrating random batch mask and multi-scale representation learning
CN112130200A (en) * 2020-09-23 2020-12-25 电子科技大学 Fault identification method based on grad-CAM attention guidance

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
MIN QIN等: "A New Improved Convolutional Neural Network Flower Image Recognition Model", 《2019 IEEE SYMPOSIUM SERIES ON COMPUTATIONAL INTELLIGENCE (SSCI)》 *
廖南星等: "基于类激活映射-注意力机制的图像描述方法", 《山东大学学报(工学版)》 *
董潇潇: "基于注意力掩模融合的目标检测算法", 《液晶与显示》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113256636A (en) * 2021-07-15 2021-08-13 北京小蝇科技有限责任公司 Bottom-up parasite species development stage and image pixel classification method
WO2023284340A1 (en) * 2021-07-15 2023-01-19 北京小蝇科技有限责任公司 Method for classifying species and development stage of parasite and classifying image pixel from bottom to top

Also Published As

Publication number Publication date
CN112749667B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
CN109949317B (en) Semi-supervised image example segmentation method based on gradual confrontation learning
CN110443818B (en) Graffiti-based weak supervision semantic segmentation method and system
CN109272500B (en) Fabric classification method based on adaptive convolutional neural network
CN106971155B (en) Unmanned vehicle lane scene segmentation method based on height information
CN111639564B (en) Video pedestrian re-identification method based on multi-attention heterogeneous network
CN113076994B (en) Open-set domain self-adaptive image classification method and system
CN113033454B (en) Method for detecting building change in urban video shooting
CN111079847B (en) Remote sensing image automatic labeling method based on deep learning
CN111428556A (en) Traffic sign recognition method based on capsule neural network
CN110210433B (en) Container number detection and identification method based on deep learning
CN112801182B (en) RGBT target tracking method based on difficult sample perception
CN113139594B (en) Self-adaptive detection method for airborne image unmanned aerial vehicle target
CN113486886B (en) License plate recognition method and device in natural scene
CN115690541A (en) Deep learning training method for improving recognition accuracy of small sample and small target
CN115063832A (en) Global and local feature-based cross-modal pedestrian re-identification method for counterstudy
CN112233105A (en) Road crack detection method based on improved FCN
CN112749667B (en) Deep learning-based nematode classification and identification method
CN111310820A (en) Foundation meteorological cloud chart classification method based on cross validation depth CNN feature integration
CN114463340A (en) Edge information guided agile remote sensing image semantic segmentation method
CN109255794B (en) Standard part depth full convolution characteristic edge detection method
CN115205614A (en) Ore X-ray image identification method for intelligent manufacturing
CN114092827A (en) Image data set generation method
CN114445662A (en) Robust image classification method and system based on label embedding
CN113591610A (en) Crop leaf aphid detection method based on computer vision
CN116229381B (en) River and lake sand production ship face recognition method

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
GR01 Patent grant
GR01 Patent grant