一种局部线性编码的深度学习果蔬分类算法

Deep Learning Fruit and Vegetable Classification Algorithm Based on Local Linear Coding

传统的图像分类问题多使用人工设计的图像特征进行分类,然而部分果蔬图像存在颜色、纹理和形状差异较小的现象,导致传统特征分类效果不够理想.针对这一问题,本文提出一种融合人工特征和深度学习特征的果蔬分类算法.首先使用Inception V3预训练模型提取果蔬图像的卷积神经网络特征;其次提取图像的颜色直方图和SIFT特征,并对SIFT特征进行局部线性编码;接着使用判别相关分析对特征进行降维融合;最后使用SVM进行训练得到分类器.通过自建果蔬图像数据库下的试验结果表明:DCA降维融合后的特征在果蔬分类准确性和速度上明显优于原特征,识别率达到近97%,更适合果蔬分类.

Traditional image classification mostly uses artificial image features for classification.However,some fruit and vegetable images have small differences in color,texture and shape,which leads to the unsatisfactory classification effect of traditional features.In order to solve this problem,this paper proposed a fruit and vegetable classification algorithm combining artificial features and Deep learning features.Firstly,Inception V3 pre-training model was used to extract the convolution features of fruit and vegetable images.Secondly,color histogram and SIFT feature of image were extracted,and local linear coding of SIFT feature was performed.Then using discriminant correlation analysis for feature reduction fusion;Finally,SVM was used to train the classifier.The experiment results of the self-built fruit and vegetable image database show that the dimension-reduction fusion feature is better than the original feature in classification ability and speed,and the recognition rate is nearly 97%,which is more suitable for fruit and vegetable classification.