A High-similarity shellfish recognition method based on convolutional neural network

IF 7.7 Q1 AGRICULTURE, MULTIDISCIPLINARY

Information Processing in Agriculture Pub Date : 2023-06-01 DOI:10.1016/j.inpa.2022.05.009

Yang Zhang , Jun Yue , Aihuan Song , Shixiang Jia , Zhenbo Li

{"title":"A High-similarity shellfish recognition method based on convolutional neural network","authors":"Yang Zhang , Jun Yue , Aihuan Song , Shixiang Jia , Zhenbo Li","doi":"10.1016/j.inpa.2022.05.009","DOIUrl":null,"url":null,"abstract":"<div><p>The high similarity of shellfish images and unbalanced samples are key factors affecting the accuracy of shellfish recognition. This study proposes a new shellfish recognition method FL_Net based on a Convolutional Neural Network (CNN). We first establish the shellfish image (SI) dataset with 68 species and 93 574 images, and then propose a filter pruning and repairing model driven by an output entropy and orthogonality measurement for the recognition of shellfish with high similarity features to improve the feature expression ability of valid information. For the shellfish recognition with unbalanced samples, a hybrid loss function, including regularization term and focus loss term, is employed to reduce the weight of easily classified samples by controlling the shared weight of each sample species to the total loss. The experimental results show that the accuracy of shellfish recognition of the proposed method is 93.95%, 13.68% higher than the benchmark network (VGG16), and the accuracy of shellfish recognition is improved by 0.46%, 17.41%, 17.36%, 4.46%, 1.67%, and 1.03% respectively compared with AlexNet, GoogLeNet, ResNet50, SN_Net, MutualNet, and ResNeSt, which are used to verify the efficiency of the proposed method.</p></div>","PeriodicalId":53443,"journal":{"name":"Information Processing in Agriculture","volume":"10 2","pages":"Pages 149-163"},"PeriodicalIF":7.7000,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Information Processing in Agriculture","FirstCategoryId":"1091","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2214317322000555","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"AGRICULTURE, MULTIDISCIPLINARY","Score":null,"Total":0}

引用次数: 0

Abstract

The high similarity of shellfish images and unbalanced samples are key factors affecting the accuracy of shellfish recognition. This study proposes a new shellfish recognition method FL_Net based on a Convolutional Neural Network (CNN). We first establish the shellfish image (SI) dataset with 68 species and 93 574 images, and then propose a filter pruning and repairing model driven by an output entropy and orthogonality measurement for the recognition of shellfish with high similarity features to improve the feature expression ability of valid information. For the shellfish recognition with unbalanced samples, a hybrid loss function, including regularization term and focus loss term, is employed to reduce the weight of easily classified samples by controlling the shared weight of each sample species to the total loss. The experimental results show that the accuracy of shellfish recognition of the proposed method is 93.95%, 13.68% higher than the benchmark network (VGG16), and the accuracy of shellfish recognition is improved by 0.46%, 17.41%, 17.36%, 4.46%, 1.67%, and 1.03% respectively compared with AlexNet, GoogLeNet, ResNet50, SN_Net, MutualNet, and ResNeSt, which are used to verify the efficiency of the proposed method.

查看原文本刊更多论文

一种基于卷积神经网络的高相似度贝类识别方法

贝类图像的高相似性和样本的不平衡是影响贝类识别精度的关键因素。本研究提出了一种基于卷积神经网络(CNN)的贝类识别新方法FL_Net。首先建立了包含68个物种、93 574幅图像的贝类图像数据集，然后提出了一种基于输出熵和正交度量驱动的滤波器剪枝修复模型，用于识别具有高相似性特征的贝类，以提高有效信息的特征表达能力。对于样本不平衡的贝类识别，采用包含正则化项和焦点损失项的混合损失函数，通过控制各样本物种的共享权重到总损失，降低易分类样本的权重。实验结果表明，本文方法的贝类识别准确率比基准网络(VGG16)提高了93.95%、13.68%，与AlexNet、GoogLeNet、ResNet50、SN_Net、MutualNet和ResNeSt相比，贝类识别准确率分别提高了0.46%、17.41%、17.36%、4.46%、1.67%和1.03%，验证了本文方法的有效性。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Information Processing in Agriculture Agricultural and Biological Sciences-Animal Science and Zoology

CiteScore

21.10

自引率

0.00%

发文量

期刊介绍： Information Processing in Agriculture (IPA) was established in 2013 and it encourages the development towards a science and technology of information processing in agriculture, through the following aims: • Promote the use of knowledge and methods from the information processing technologies in the agriculture; • Illustrate the experiences and publications of the institutes, universities and government, and also the profitable technologies on agriculture; • Provide opportunities and platform for exchanging knowledge, strategies and experiences among the researchers in information processing worldwide; • Promote and encourage interactions among agriculture Scientists, Meteorologists, Biologists (Pathologists/Entomologists) with IT Professionals and other stakeholders to develop and implement methods, techniques, tools, and issues related to information processing technology in agriculture; • Create and promote expert groups for development of agro-meteorological databases, crop and livestock modelling and applications for development of crop performance based decision support system. Topics of interest include, but are not limited to: • Smart Sensor and Wireless Sensor Network • Remote Sensing • Simulation, Optimization, Modeling and Automatic Control • Decision Support Systems, Intelligent Systems and Artificial Intelligence • Computer Vision and Image Processing • Inspection and Traceability for Food Quality • Precision Agriculture and Intelligent Instrument • The Internet of Things and Cloud Computing • Big Data and Data Mining