基于改进MobileNet轻量级网络的水稻病害识别  

Rice Disease Recognition Based on Improved MobileNet Lightweight Network

作　　者：郑超 张华民 ZHENG Chao;ZHANG Huamin(College of Mechanical Engineering,Anhui Science and Technology University,Fengyang 233100,China)

机构地区：[1]安徽科技学院机械工程学院,安徽凤阳233100

出　　处：《荆楚理工学院学报》2024年第4期26-32,共7页Journal of Jingchu University of Technology

基　　金：安徽省高校自然科学研究重点项目(KJ2019A0803);安徽省自然科学基金项目(1708085QF146)。

摘　　要：目的:针对传统MobileNet-v2模型水稻叶面病害识别过程中出现的准确率低、运行速度慢、特征提取难等问题,提出一种基于改进MobileNet-v2轻量级网络的水稻叶面病害识别模型。方法:该模型采用增加注意力机制模块的结构方法增强图像的特征提取,然后将预训练好的权重参数迁移到改进的模型中,进而对水稻4种叶面病害进行识别研究。结果:该模型在50个epoch的训练测试过程中,训练速度和过拟合问题得到了较大的改善,最终测试识别准确率较传统MobileNet-v2模型准确率提高了7.97%。结论:该模型在水稻叶面病害识别中准确率较高,识别速度较快,为水稻叶面病害的识别与研究提供了参考和借鉴意义。Objective:To address the problem of the conventional MobileNet-v2 model in the recognition of rice leaf diseases,including its low recognition accuracy,sluggish running speed,and challenging feature extraction,a rice leaf disease recognition model based on improved MobileNet-v2 lightweight network was proposed.Methods:In this model,the method of adding attention mechanism module is used to enhance image feature extraction,and then the weight parameters of pre-trained model were transferred to the improved model to identify the four leaf diseases of rice.Results:The training speed and overfitting issues were significantly reduced throughout the training and testing of the new-mobile model over 50 epochs,and the final test recognition accuracy was 7.3%higher than that of the conventional MobileNet-v2 model.Conclusions:The new mobile model is more accurate and quicker in recognizing rice leaf diseases,which provides a reference and significance for identifying and researching of rice leaf diseases.

关 键 词：图像识别 水稻病害 迁移学习 MobileNet-v2 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]