%0 Journal Article
%T 基于图像深度特征的稻种识别方法研究<br>Research on Rice Seed Recognition Method Based on Image Depth Feature
%A 黄崇渝
%A 华云松
%J Software Engineering and Applications
%P 1272-1281
%@ 2325-2278
%D 2022
%I Hans Publishing
%R 10.12677/SEA.2022.116130
%X 随着水稻品种不断增加，如何识别稻种成为业内的一个难点，机器视觉技术较人工识别稻种识别准确度和时间更具优势，因而成为一种趋势。针对传统图像特征识别方法能力有限、识别精度低、速度慢的问题，本文提出使用L2正则化和dropout的卷积神经网络(convolutional neural networks, CNN)将原来得到的线性不可分图像深度特征数据变为线性可分的数据，通过支持向量机(support vector machine, SVM)代替传统CNN的softmax分类器对得到的线性数据进行分类，最后利用10个常用CNN模型 + SVM，通过准确性、敏感性、特异性、假阳性率、F1 Score和测试时间等方面对模型效果进行评估后，得出水稻识别结果。实验表明，该方法在稻种识别方面准确率达到99.8%，每张图片识别速度低至0.07 s，可用于混杂稻种的识别和分类。<br />
With the increasing number of rice varieties, how to identify rice seeds has become a difficult point in the industry. Machine vision technology has become a trend as it has more advantages over manual identification of rice seeds in terms of accuracy and time. To address the problems of limited capability, low recognition accuracy and slow speed of traditional image feature recognition methods, in this paper, we propose a convolutional neural network (CNN) using L2 regularization and dropout to change the originally obtained linear non-separable image depth feature data into linear separable data by support vector machine (SVM) instead of traditional. Finally, the rice recognition results were obtained by evaluating the model effects in terms of accuracy, sensitivity, specificity, false positive rate, F1 Score and testing time using 10 commonly used CNN models + SVM. Experiments show that the method achieves 99.8% accuracy in rice seed recognition, and the recognition speed is as low as 0.07 s per image, which can be used for the recognition and classification of mixed rice seeds.
%K 稻种识别，深度学习，卷积神经网络，支持向量机<br>Rice Seed Identification
%K Deep Learning
%K Convolutional Neural Networks
%K Support Vector Machines
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=59183