基于EfficientNetV2的玉米粒外观破损检测方法  

A Corn Kernel Quality Detection Method Based on the EfficientNetV2

作　　者：陈泊洁 CHEN Bojie(College of Mechanical Engineering,Yangzhou University,Yangzhou 225009,China)

机构地区：[1]扬州大学机械工程学院,江苏扬州225009

出　　处：《现代农业装备》2025年第1期56-61,74,共7页Modern Agricultural Equipment

摘　　要：针对传统玉米粒外观破损检测方法存在的效率低下和准确性不足的问题,提出了一种基于EfficientNetV2模型的检测方法,旨在实现对玉米粒外观破损的高效、精确检测。首先,搭建了图像采集装置,构建了一个包含完好及破损玉米粒图像的高质量数据集;然后,对这些图像进行了细致的预处理和精确的图像分割,以增强特征的可识别性;最后,利用EfficientNetV2模型对数据集进行了系统训练。试验结果显示,模型在训练集上的准确率稳步提升,最终达到99.76%,并且在后期大多数epoch中验证集的准确率达到了100%;此外,每张图像切片的处理时间仅为0.028 s,这一速度足以支持实时检测的需求。In response to the issues of low efficiency and insufficient accuracy in traditional methods for detecting damage to corn kernels,this paper proposed a detection method based on the EfficientNetV2 model for the purpose of achieving efficient and precise detection of corn kernel damage.Initially,an image acquisition device was constructed,and a high-quality dataset containing images of both intact and damaged corn kernels was built.These images then underwent meticulous preprocessing and precise image segmentation to enhance feature recognizability.Subsequently,the EfficientNetV2 model was systematically trained using this dataset.Experimental results indicated that the model's accuracy on the training set steadily increased,ultimately reaching 99.76%.On the validation set,the accuracy reached 100%in the majority of later epochs.Additionally,the processing time for each image slice was only 0.028 seconds,which was sufficient to support real-time detection requirements.

关 键 词：玉米粒外观破损检测 EfficientNetV2 深度学习 图像识别 

分 类 号：S24[农业科学—农业电气化与自动化] TP391.41[农业科学—农业工程]