液晶与显示, 2019, 34 (12): 1202, 网络出版: 2020-01-09
基于改进深度残差网络的河蟹精准溯源系统
Accurate traceability system of crab based on improved deep residual network
特征提取 深度残差网络 河蟹识别 河蟹溯源 feature extraction deep residual network crab recognition traceability of crab
摘要
针对当前商品河蟹质量安全问题逐渐增多、真假阳澄湖大闸蟹难辨、深度残差网络提取特征维度大的问题, 提出一种基于改进深度残差网络的河蟹精准溯源系统。系统由养殖环节、检测环节、销售环节、溯源环节4部分组成, 养殖、检测、销售环节将河蟹的养殖、检测、销售数据保存到溯源数据库, 溯源环节通过基于改进深度残差网络的河蟹识别技术, 识别溯源数据库中是否存在待溯源河蟹, 并根据识别结果输出待溯源河蟹在养殖、检测、销售环节的数据, 最终实现每一只商品河蟹从消费者到养殖场的精准溯源追踪。实验结果表明, 改进深度残差网络的河蟹识别技术将提取的蟹壳特征向量从2 048维降至156维, 识别耗时降低了92%, 识别准确率为92.1%。
Abstract
Aiming at the problems of current commercial river crab quality safety, hard to distinguish between real and false Yangcheng Lake hairy crabs, and too large feature dimension of deep residual network extraction, a precise source tracing system of river crab based on improved deep residual network is proposed. The system consists of four parts: breed, detection, sale and traceability. In breed, detection and sale parts, the data of crab breed, detection and sale are stored in traceability database. The traceability part identifies whether there is traceable river in traceability database by identifying crab based on improved deep residual network. According to the result of identification, the data of crab breeding, testing and marketing are output. Finally, the precise traceability of each commercial crab from consumer to farm is realized. The improved crab identification technology of the deep residual network can reduced the extracted crab shell feature vector from 2 048 to 156 dimensions, the recognition time is reduced by 92%, and the recognition accuracy is 92.1%.
侍国忠, 陈明, 张重阳. 基于改进深度残差网络的河蟹精准溯源系统[J]. 液晶与显示, 2019, 34(12): 1202. SHI Guo-zhong, CHEN Ming, ZHANG Chong-yang. Accurate traceability system of crab based on improved deep residual network[J]. Chinese Journal of Liquid Crystals and Displays, 2019, 34(12): 1202.