基于改进Mask R-CNN的苹果园害虫识别方法  

Pest Identification Method in Apple Orchard Based on Improved Mask R-CNN

作　　者：王金星[1,2] 马博 王震 刘双喜[1,4] 慕君林 王云飞[1] WANG Jinxing;MA Bo;WANG Zhen;LIU Shuangxi;MU Junlin;WANG Yunfei(College of Mechanical and Electronic Engineering,Shandong Agricultural University,Taian 271018,China;Shandong Agricultural Equipment Intelligent Engineering Laboratory,Taian 271018,China;College of Horticulture Science and Engineering,Shandong Agricultural University,Taian 271018,China;Shandong Provincial Key Laboratory of Horticultural Machinery and Equipment,Taian 271018,China)

机构地区：[1]山东农业大学机械与电子工程学院,泰安271018 [2]山东省农业装备智能化工程实验室,泰安271018 [3]山东农业大学园艺科学与工程学院,泰安271018 [4]山东省园艺机械与装备重点实验室,泰安271018

出　　处：《农业机械学报》2023年第6期253-263,360,共12页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金项目(32071908);财政部和农业农村部:国家现代农业(苹果)产业技术体系项目(CARS-27)。

摘　　要：针对基础卷积神经网络识别苹果园害虫易受背景干扰及重要特征表达能力不强问题,提出一种基于改进Mask R-CNN的苹果园害虫识别方法。首先,基于Haar特征方法对多点采集得到的苹果园害虫图像进行迭代初分割,提取害虫单体图像样本,并对该样本进行多途径扩增,得到用于深度学习的扩增样本数据集。其次,对Mask R-CNN中的特征提取网络进行优化,采用嵌入注意力机制模块CBAM的ResNeXt网络作为改进模型的Backbone,增加模型对害虫空间及语义信息的提取,有效避免背景对模型性能的影响;同时引入Boundary损失函数,避免害虫掩膜边缘缺失及定位不准确问题。最后,以原始Mask R-CNN模型作为对照模型,平均精度均值作为评价指标进行试验。结果表明,改进Mask R-CNN模型平均精度均值达到96.52%,相比于原始Mask R-CNN模型,提高4.21个百分点,改进Mask R-CNN可精准有效识别苹果园害虫,为苹果园病虫害绿色防控提供技术支持。Aiming at the problem that the basic convolutional neural network is vulnerable to background interference and the expression ability of important features is not strong in apple orchard pest recognition,an apple orchard pest recognition method based on improved Mask R CNN was proposed.Firstly,based on Haar feature method,the apple orchard pest images collected from multiple points were iteratively preliminarily segmented,the single pest image sample was extracted,and multi⁃channel amplification on the sample was performed to obtain the amplified sample data for deep learning.Secondly,the feature extraction network in Mask R CNN was optimized,and the ResNeXt network embedded in the attention mechanism module CBAM was used as the Backbone of the improved model,which increased the extraction of pest space and semantic information by the model,and effectively avoided the influence of background on performance of the model.At the same time,the Boundary loss function was introduced to avoid the problem of missing edge of pest mask and inaccurate positioning.Finally,the original Mask R CNN model was used as the control model,and the mean average precision(mAP)was used as the evaluation index to conduct experiments.The results showed that the mean average precision of the improved Mask R CNN model reached 96.52%.Compared with the original Mask R CNN model,the mean average precision was increased by 4.21 percentage points.The results showed that the improved Mask R CNN can accurately and effectively identify pests in apple orchards.The research result can provide technical support for green control of apple orchard pests and diseases.

关 键 词：苹果园 害虫识别 深度学习 注意力机制 损失函数 Mask R-CNN 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]