摘要
针对传统人工挑选存在误差大、效率低等问题,提出一种基于ResNet网络的细粒度图像分类方法对玉米种子进行精确分类。首先,针对玉米种子特征建立图像采集平台并对采集到的图片进行图像预处理,然后,构建基于ResNet的玉米种子的分类模型并优化整个网络,输入玉米种子图像输入到模型中进行训练,直到得到权值最好的分类模型,相同条件下还比较VGG-13、AlexNet以及ResNet-50的结果。试验表明:本文提出的卷积神经网络的识别精度和识别时间与VGG-13、AlexNet以及ResNet-50相比都得到一定的提升。经过改进ResNet算法对大、中、小种子的识别率分别为96.4%、93.5%、92.3%高于VGG-13、AlexNet以及ResNet-50三种算法。
Aiming at the problems of high error and low efficiency in the traditional manual selection,a fine-grained image classification method based on the ResNet network was proposed to accurately classify corn seeds.First,an image acquisition platform for corn seed characteristics was established,and collected pictures were processed.Then,a ResNet-based corn seed classification model was constructed and optimized.Inputting corn seed images into the model for training continuously until get the classification model with the best weight value also compares the results of VGG-13,AlexNet and ResNet-50 under the same conditions.The experimental results showed that the recognition accuracy and the time of the convolutional neural network proposed in this paper were improved compared with VGG-13,AlexNet,and RESnet-50.The recognition rates of the modified ResNet algorithm for large,medium,and small seeds were 96.4%,93.5%,and 92.3%,respectively,which were higher than VGG-13,AlexNet,and RESnet-50.
作者
吕梦棋
张芮祥
贾浩
马丽
Lü Mengqi;Zhang Ruixiang;Jia Hao;Ma Li(College of Information Technology,Jilin Agricultural University,Changchun,130118,China)
出处
《中国农机化学报》
北大核心
2021年第4期92-98,共7页
Journal of Chinese Agricultural Mechanization
基金
国家自然科学基金-联合基金(U19A2061)。
