CN114677672A - Mature blueberry fruit identification method based on deep learning neural network - Google Patents

Mature blueberry fruit identification method based on deep learning neural network Download PDF

Info

Publication number
CN114677672A
CN114677672A CN202210307641.6A CN202210307641A CN114677672A CN 114677672 A CN114677672 A CN 114677672A CN 202210307641 A CN202210307641 A CN 202210307641A CN 114677672 A CN114677672 A CN 114677672A
Authority
CN
China
Prior art keywords
neural network
deep learning
training
mature
network
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
Application number
CN202210307641.6A
Other languages
Chinese (zh)
Inventor
闫祥海
陈炳鑫
徐立友
张静云
吴依伟
赵文正
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.)
Henan University of Science and Technology
Original Assignee
Henan University of Science and Technology
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 Henan University of Science and Technology filed Critical Henan University of Science and Technology
Priority to CN202210307641.6A priority Critical patent/CN114677672A/en
Publication of CN114677672A publication Critical patent/CN114677672A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • 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

Landscapes

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

Abstract

The invention discloses a mature blueberry fruit identification method based on a deep learning neural network, which is used for training by combining a ResNet50 residual neural network and a YOLOv2 detection network, and specifically comprises the following steps: preprocessing the collected blueberry fruit picture set, and augmenting and expanding the data set by using data; proportionally dividing the augmented data set into a training set, a test set and a verification set; labeling image features; extracting image characteristics by adopting a ResNet50 residual error network; inputting the extracted features as an entry function into a YOLOv2 detection network for target detection; the identification accuracy rate of the concentrated mature blueberry fruits is 95% through testing, and the concentrated mature blueberry fruits are verified to be correctly identified. The method skillfully combines the advantages of the ResNet50 residual neural network and the YOlOv2 detection network, the ResNet50 residual neural network effectively solves the problems of gradient explosion and gradient disappearance caused by deepening of the layer number, the YOv 2 detection network is accurate in prediction and high in speed, and the training precision and the recognition reaction speed of the neural network are enhanced by combining the ResNet50 residual neural network and the YOlOv2 detection network.

Description

Mature blueberry fruit identification method based on deep learning neural network
Technical Field
The invention relates to the technical field of computer vision, in particular to a mature blueberry fruit identification method based on a deep learning neural network.
Background
At the present stage, the blueberry maturity information is acquired mainly by means of manual observation, observers perform field actual measurement on the blueberries according to the definition and description in the agricultural meteorological observation standard, the efficiency is low, a large amount of manpower and material resources need to be consumed, and the real-time and rapid monitoring requirements cannot be met. Nowadays, intelligent recognition technology is more and more used in the agricultural field, and contemporary computers have superstrong operational capability, input picture data to neural network, through abstract data learning, train out the neural network that has the specific crops of discernment classification to improve production efficiency.
Disclosure of Invention
In view of the foregoing background, the present invention is directed to a method for identifying ripe blueberry fruits based on a deep learning neural network. According to the method, the ResNet50 residual neural network is used for extracting features, the YOLOv2 detection network is used for carrying out feature recognition training, and the deep learning neural network obtained by combining the two networks has the capability of recognizing mature blueberry fruits and can keep high precision.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a mature blueberry fruit identification method based on a deep learning neural network comprises the following specific implementation steps:
step 1, preprocessing a collected blueberry fruit picture set, and expanding a data set through data augmentation operation;
step 2, proportionally dividing the augmented image data set into a training set, a test set and a verification set;
step 3, labeling image characteristics: labeling the mature blueberry fruits in each picture in the training set;
step 4, constructing a deep learning neural network: and performing image feature extraction by using a ResNet50 residual neural network. Taking the training set labeled by the image characteristics in the step 3 as an input image, and extracting the deep learning characteristics of the mature blueberry fruits from 40-relu in a ResNet50 residual neural network;
step 5, inputting the features extracted in the step 4 as an entry function into a YOLOV2 detection network for target detection training to obtain a deep learning neural network model;
step 6, adopting the deep learning neural network model trained in the step 5 to perform target recognition on the test set and the verification set, wherein the training result is as follows: the identification accuracy rate of the concentrated mature blueberry fruits is 95% through testing, and the concentrated mature blueberry fruits are verified to be correctly identified.
The data augmentation in the step 1 is to augment the image by using the following 4 functions to increase the training amount and enhance the training effect:
mirror image turning: the left and right turning images do not change the types of objects generally, and are shot around different sides of blueberry plants in a simulated mode;
adding illumination: different sun lights are added, and the blueberry plant patterns under different lighting conditions are simulated;
random segmentation: in order to reduce the sensitivity of the convolutional layer to the target position, the mature blueberry fruits appear at different positions of the picture in different proportions through image random cropping.
Picture rotation: and rotating the picture by 90 degrees, 180 degrees and 270 degrees to obtain the blueberry plants in different field orientations.
In the step 2, the image data set division is to randomly divide the picture data into a training set and a testing set and a verification set according to the proportion of 70%, 20% and 10% under a caffe framework of a windows system.
The step 3 comprises the following steps: and manually labeling mature blueberry fruits in each picture in the training set to form a deep learning mature blueberry fruit picture database.
And step 4, extracting picture features for the next step. As shown in fig. 2, the ResNet50 residual neural network is divided into 177 layers of 5 stages. Where Stage0 is the pre-processing of the input signal, the last 4 stages are all composed of the bottleneck layer. The input signal is a training set labeled by the image characteristics in the step 3, and the training set is converted into a format of 1mdb and used as the input of a ResNet residual error neural network model; wherein the convolutional neural network is used for extracting features, and the mature blueberry fruit deep learning features extracted from the layer 141, 40_ relu, will be used as an entry function of YOLOV 2.
And step 5, inputting the features extracted in the step 4 as an entry function into a YOLOV2 detection network for target detection training, wherein the YOLOV2 detection can effectively reduce model overfitting.
And 6, completing training, wherein the training result is that the recognition accuracy rate of the concentrated mature blueberry fruits is 95% in the test, and the mature blueberry fruits can be correctly recognized by using the neural network model.
The invention skillfully combines the advantages of the ResNet50 residual neural network and the YOlOv2 detection network. The ResNet50 residual neural network does a reference (X) to the input of each layer, and learns to form a residual function, rather than learning some functions without the reference (X), the residual function is easier to optimize, the number of layers of the network is greatly deepened, the network structure is networked and three-dimensional, and simultaneously the residual neural network effectively solves the problems of gradient explosion and gradient disappearance caused by the deepening of the number of layers. The YOLOv2 detection network has accurate prediction and high speed. The combination of the two enhances the training precision and the recognition response speed of the neural network.
The training of the neural network only marks blueberry fruit pictures which can be directly picked by a picking machine and have high maturity. The method has the advantages that blueberry fruits which cannot be directly picked by a machine and have the exposure degree lower than 50%, fruits with low maturity, bad fruits and the like are selectively not marked. The labeling method reduces the picking difficulty of a machine and ensures the quality and the maturity of blueberry fruits. The invention is combined with a blueberry picking machine, can grasp the optimal picking time, picks high-quality fruits in batches in a large range in real time, and is of practical help to the industrial production of blueberry crops.
Drawings
Fig. 1 is a flow architecture diagram of a mature blueberry fruit identification method based on a deep learning neural network.
Fig. 2 is a schematic diagram of a specific structure of each stage in the ResNet50 residual neural network adopted in the present invention.
Fig. 3 is a schematic diagram of a YOLOv2 detection network structure adopted in the present invention.
FIG. 4 is a diagram illustrating the verification result of the verification set in the present invention.
FIG. 5 is a schematic diagram of the recognition accuracy of the training model according to the present invention.
Detailed Description
The present invention is further described in detail with reference to the drawings and the specific embodiments so that the advantages and features of the present invention can be easily understood by those skilled in the art, and the scope of the present invention can be clearly and clearly defined. Referring to fig. 1, the present invention provides a method for identifying ripe blueberry fruits based on a deep learning neural network, which includes the following steps:
step 1, preprocessing a collected blueberry fruit picture set, and expanding a data set through data augmentation operation;
step 2, dividing the augmented image data set into a training set, a test set and a verification set according to a proportion;
step 3, marking image characteristics: marking all mature blueberry fruits in each picture in the training set;
step 4, constructing a deep learning neural network: and performing image feature extraction by using a ResNet50 residual neural network. Taking the training set labeled by the image characteristics in the step 3 as an input image, and extracting the deep learning characteristics of the mature blueberry fruits from the 141 th layer in the ResNet50 residual neural network;
step 5, inputting the features extracted in the step 4 into a Yolov2 detection network for target detection training by taking the features as an entry function to obtain a deep learning neural network model;
step 6, adopting the deep learning neural network model trained in the step 5 to perform target recognition on the test set and the verification set, wherein the training result is as follows: the identification accuracy rate of the concentrated mature blueberry fruits is 95% through testing, and the concentrated mature blueberry fruits are verified to be correctly identified.
In step 1, the data set expansion uses the following 4 functions to augment the data:
mirrorring: mirror image turning is carried out, shooting angles on different sides are simulated, and pictures are enlarged by 2 times;
color shifting: simulating sunlight change for RGB different channel values through addition and subtraction coefficients;
croping: randomly cutting to reduce the sensitivity of the convolutional layer to a target position, and randomly cutting the image to enable the mature blueberry fruits to appear at different positions of the picture in different proportions;
picture rotation function: the picture is rotated by 90 degrees, 180 degrees and 270 degrees around the pixel center point, and the picture is enlarged by 3 times.
In the step 2, the image data set division is to randomly divide the picture data into a training set and a testing set and a verification set according to the proportion of 70%, 20% and 10% under a caffe framework of a windows system.
In the step 3, the mature blueberry fruits in each picture in the training set need to be labeled manually to form a deep learning mature blueberry fruit picture database.
And in the step 4, the ResNet50 residual neural network is adopted to extract picture features. Referring to fig. 2, the ResNet50 residual network is divided into 5 stages, for 177 layers. Wherein Stage0 is the preprocessing of the input signal, the input signal is the training set labeled by the image characteristics in step 3, the training set is converted into 1mdb format and used as the input of ResNet50 residual neural network model. The last 4 stages (i.e. Stage1, Stage2, Stage3, Stage4) all consist of bottleneck layers, convolutional neural networks are used to extract features, and the mature blueberry fruit deep learning features extracted by 40_ relu at layer 141 will be used as the entry function of YOLOV 2.
And in the step 5, the mature blueberry fruit deep learning characteristics extracted in the step 4 are used as an entry function and input into a Yolov2 detection network for target detection training, so that model overfitting can be effectively reduced through Yolov2 detection.
Referring to fig. 3, the YOLOV2 test network uses a new basic model, including 19 convolutional layers and 5 max pooling layers, and after using 3 × 3 convolution and 2 × 2 max pooling layers, the feature dimension decreases by two times while the number of channels in the feature map increases by two times. The prediction is performed by using global average pooling, and 1 × 1 convolution is used between 3 × 3 convolution to compress the channel number of the feature map so as to reduce the model calculation amount and parameters. Depth layers are also used behind each convolution layer to speed up convergence and reduce the degree of model overfitting.
And step 6 is a result schematic diagram of target recognition of the test set and the verification set by adopting the deep learning neural network model trained in the step 5. Referring to fig. 4 and 5, the training result shows that the recognition accuracy of concentrated mature blueberry fruits can reach 95%, and the mature blueberry fruits can be correctly recognized by using the neural network model provided by the invention.
The above description is only a preferred embodiment of the present invention, and any person skilled in the art can make any simple modification, equivalent change and modification to the above embodiments according to the technical essence of the present invention without departing from the scope of the present invention, and still fall within the scope of the present invention.

Claims (5)

1. A mature blueberry fruit identification method based on a deep learning neural network is characterized by comprising the following steps:
step 1, preprocessing a collected blueberry fruit picture set, and expanding a data set through data augmentation operation;
step 2, dividing the augmented image data set into a training set, a test set and a verification set according to a proportion;
step 3, marking image characteristics: labeling the mature blueberry fruits in each picture in the training set;
step 4, constructing a deep learning neural network: taking the training set after the image features are labeled in the step 3 as an input signal, and adopting a ResNet50 residual error network to extract the image features;
step 5, inputting the features extracted in the step 4 as an entry function into a YOLOV2 detection network for target detection training to obtain a deep learning neural network model;
step 6, adopting the deep learning neural network model trained in the step 5 to perform target recognition on the test set and the verification set, wherein the training result is as follows: the identification accuracy rate of the concentrated mature blueberry fruits is 95% through testing, and the concentrated mature blueberry fruits are verified to be correctly identified.
2. The method as claimed in claim 1, wherein in step 1, the data augmentation operation includes mirror image flipping, illumination adding, random segmentation, and image rotation.
3. The method for identifying ripe blueberry fruits based on the deep learning neural network as claimed in claim 1, wherein in the step 2, the image data set division is that the image data is divided into a training set, a testing set and a verification set at random according to proportions of 70%, 20% and 10% under a caffe frame of a windows system.
4. The method for identifying the mature blueberry fruits based on the deep learning neural network as claimed in claim 1, wherein in the step 3, only blueberry fruit pictures which can meet the direct picking requirement of a picking machine and have high maturity are labeled when the image features are labeled.
5. The method for identifying ripe blueberry fruits based on the deep learning neural network as claimed in claim 1, wherein the specific process of the step 4 is as follows: the ResNet50 residual neural network is divided into 5 stages, wherein Stage0 preprocesses input signals, Stage1, Stage2, Stage3 and Stage4 are all composed of bottleneck layers, the input signals are training sets labeled by the image features in the step 3, and the training sets are converted into a 1mdb format and used as the input of a ResNet50 residual neural network model; the ResNet50 residual neural network has 177 layers, wherein the convolutional neural network is used for extracting features, and the mature blueberry fruit deep learning features extracted by 40_ relu at the 141 layer are used as the entry function of the YOLOV2 detection network.
CN202210307641.6A 2022-03-27 2022-03-27 Mature blueberry fruit identification method based on deep learning neural network Pending CN114677672A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210307641.6A CN114677672A (en) 2022-03-27 2022-03-27 Mature blueberry fruit identification method based on deep learning neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210307641.6A CN114677672A (en) 2022-03-27 2022-03-27 Mature blueberry fruit identification method based on deep learning neural network

Publications (1)

Publication Number Publication Date
CN114677672A true CN114677672A (en) 2022-06-28

Family

ID=82075738

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210307641.6A Pending CN114677672A (en) 2022-03-27 2022-03-27 Mature blueberry fruit identification method based on deep learning neural network

Country Status (1)

Country Link
CN (1) CN114677672A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116958783A (en) * 2023-07-24 2023-10-27 中国矿业大学 Light-weight image recognition method based on depth residual two-dimensional random configuration network

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116958783A (en) * 2023-07-24 2023-10-27 中国矿业大学 Light-weight image recognition method based on depth residual two-dimensional random configuration network
CN116958783B (en) * 2023-07-24 2024-02-27 中国矿业大学 Light-weight image recognition method based on depth residual two-dimensional random configuration network

Similar Documents

Publication Publication Date Title
CN109543606B (en) Human face recognition method with attention mechanism
CN110348376B (en) Pedestrian real-time detection method based on neural network
CN111126325B (en) Intelligent personnel security identification statistical method based on video
CN111400536B (en) Low-cost tomato leaf disease identification method based on lightweight deep neural network
CN107392131A (en) A kind of action identification method based on skeleton nodal distance
CN111462044A (en) Greenhouse strawberry detection and maturity evaluation method based on deep learning model
CN113128335B (en) Method, system and application for detecting, classifying and finding micro-living ancient fossil image
CN111178177A (en) Cucumber disease identification method based on convolutional neural network
CN108520203A (en) Multiple target feature extracting method based on fusion adaptive more external surrounding frames and cross pond feature
CN114140665A (en) Dense small target detection method based on improved YOLOv5
CN109508756B (en) Foundation cloud classification method based on multi-cue multi-mode fusion depth network
CN115272204A (en) Bearing surface scratch detection method based on machine vision
CN116052222A (en) Cattle face recognition method for naturally collecting cattle face image
Liu et al. Development of a machine vision algorithm for recognition of peach fruit in a natural scene
CN115984698A (en) Litchi fruit growing period identification method based on improved YOLOv5
CN113298817A (en) High-accuracy semantic segmentation method for remote sensing image
CN116416613A (en) Citrus fruit identification method and system based on improved YOLO v7
CN115205626A (en) Data enhancement method applied to field of coating defect detection
CN114677672A (en) Mature blueberry fruit identification method based on deep learning neural network
CN111832508B (en) DIE _ GA-based low-illumination target detection method
CN111597939B (en) High-speed rail line nest defect detection method based on deep learning
CN112966698A (en) Freshwater fish image real-time identification method based on lightweight convolutional network
CN117079125A (en) Kiwi fruit pollination flower identification method based on improved YOLOv5
CN113077438A (en) Cell nucleus region extraction method and imaging method for multi-cell nucleus color image
CN113837063B (en) Reinforcement learning-based curling motion field analysis and auxiliary decision-making 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