基于改进YOLOv7-tiny算法的多种类不均衡样本水稻害虫检测
Detection of Rice Pests in Multi-species Unbalanced Samples Based on Improved YOLOv7-Tiny Algorithm
李鑫 1南新元1
作者信息
- 1. 新疆大学电气工程学院,新疆 乌鲁木齐 830017
- 折叠
摘要
为实现基于机器视觉的田间水稻害虫检测,本研究结合IP102 农业害虫数据集及网络资源,建立了含有 26 类标签的不均衡样本水稻害虫数据集;改进YOLOv7-tiny单阶段目标检测算法,以部分卷积PConv作为主要卷积核,结合极化自注意力机制(Polarized Self-Attention),将提取到的特征进行复杂双向多尺度特征融合,建立了适合多种类不均衡样本的水稻害虫检测模型.结果表明,在加入迁移学习和多尺度训练的条件下,改进后的YOLOv7-tiny检测算法在自建水稻害虫数据集的平均检测精度达到 96.4%,单张图片的检测时间为 8.8 ms,模型大小为 9 055 kb,可实现对田间水稻害虫的快速准确识别,为水稻害虫的智能化检测和防治提供了技术支持.
Abstract
This study aimed to realize the detection of rice pests in field based on machine vision.Com-bined with the IP102 agricultural pest dataset and network resources,an unbalanced rice pest data set contai-ning 26 kinds of labels was established in this study.Through improving YOLOv7-tiny single-stage target de-tection algorithm,using partial convolution as the main convolution kernel,combining with the Polarized Self-Attention mechanism,and carrying out complex bidirectional multi-scale feature fusion for the extracted fea-tures,a rice pest detection model suitable for multi-species unbalanced samples was established.The results showed that under the conditions of adding transfer learning and multi-scale training,the average detection ac-curacy of the improved YOLOv7-tiny detection algorithm in the self-built rice pest data set was 96.4%,the de-tection time of single image was 8.8 ms,and the model size was 9 055 kb,so the rapid and accurate identifi-cation of rice pests in field could be realized.This study could provide technical supports for the intelligent de-tection and control of rice pests.
关键词
水稻害虫检测/改进YOLOv7-tiny算法/部分卷积/极化自注意力机制/特征融合/迁移学习Key words
Rice pest detection/Improved YOLOv7-tiny algorithm/Partial convolution/Polarized self-attention/Feature fusion/Transfer learning引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据