基于ResNet的害虫图像质量评估方法  

Research on forest pest identification method based on face image quality assessment

作　　者：王红梅[1] 朱莉 WANG Hongmei;ZHU Li(School of Computer Science&Engineering,Changchun University of Technology,Changchun 130102,China)

机构地区：[1]长春工业大学计算机科学与工程学院,吉林长春130102

出　　处：《长春工业大学学报》2024年第1期52-58,共7页Journal of Changchun University of Technology

基　　金：吉林省教育厅基金资助项目(JJKH20230767KJ)。

摘　　要：提出一种基于ResNet的害虫图像质量评估方法,从而对林业害虫图像进行预评估。该方法首先提取害虫图像特征,并通过Wasserstein距离计算不同图像特征间的相似分布距离作为质量伪标签进行训练。通过预评估区分出不同质量的林业害虫图像,对其进行筛选、识别、分类,从而达到提高识别准确率的效果。实验结果表明,经过该方法筛选后的林业害虫数据集在ResNet18和ResNet50网络上识别准确率分别提升2.97%,2.57%。To solve the above problems,we propose a ResNet-based quality assessment method for pest images to pre-evaluate forestry pest images.The method first extracts pest image features and calculates the similar distribution distance between different image features as quality pseudo-labels for training by Wasserstein distance.Then,different quality forestry pest images are distinguished by pre-evaluation and screened for recognition classification to improve recognition accuracy.The experimental results show that the recognition accuracy of the forest pest dataset screened by this method is improved by 2.97%and 2.57%on the ResNet18 and the ResNet50 networks,respectively.

关 键 词：ResNet 卷积神经网络 林业害虫 质量评估 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]