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
王红梅 1朱莉1
作者信息
- 1. 长春工业大学 计算机科学与工程学院,吉林 长春 130102
- 折叠
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%.
Abstract
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Key words
ResNet/convolutional neural network/forestry pests/quality assessment引用本文复制引用
基金项目
吉林省教育厅项目(JJKH20230767KJ)
出版年
2024
NETL
NSTL
万方数据