农机化研究2025，Vol.47Issue(9)：1-8,8.DOI:10.13427/j.issn.1003-188X.2025.09.001
基于深度学习的玉米籽粒含杂检测方法研究
Detection Method of Maize Kernel Impurity Based on Deep Learning
摘要
Abstract
The impurity rate of maize kernels is a crucial parameter for kernel harvesters.However,China's current maize kernel harvesting process lacks the capability for online real-time detection of impurity rates.To achieve online real-time detection of maize kernel impurity rates,initially devised a maize kernel impurity detection apparatus and for-mulated an impurity calculation equation.Subsequently,a dataset of maize kernel-impurity images was amassed and subjected to image enhancement and annotation.Moreover,an enhanced DeepLab v3+model,couple was proposed,coupled with a contraction-excitation-attention module,a comparative assessment involving various models was per-formed,along with field tests.The test results demonstrated that the MIoU and PA of improved DeepLab v3+deep learn-ing model attained,92.17%and 96.45%and its segmentation performance surpassed that of other models in compari-son.The augmented DeepLab v3+model employing Xception as the backbone network outperformed ResNet50 and Res-Net101.In the field test,the device-based detection results closely align with manual detection,displaying minimal ab-solute and relative errors in both cases.Thus,the impurity detection device and algorithm proposed a realistic online im-purity rate detection solution,heightening the efficiency and quality of maize kernel harvesting,thereby fostering the au-tomation of maize kernel harvesters in China.关键词
玉米收获/深度学习/含杂检测/DeepLab v3+/田间试验Key words
maize harvesting/deep learning/impurity detection/DeepLab v3+/field test分类
农业工程引用本文复制引用
杜岳峰,侯思余,李国润..基于深度学习的玉米籽粒含杂检测方法研究[J].农机化研究,2025,47(9):1-8,8.基金项目
国家自然科学基金项目(52175258) （52175258）
