基于微型计算机和神经网络的果蔬识别

Vegetable and Fruit Recognition Based on Micro Processor and Neural Network

在微型计算机上使用深度卷积神经网络实现果蔬自动识别的复杂图像识别任务,并构造了具有人工智能技术的电子秤系统.采用深度可分离卷积和反向残差网络设计构成的深度神经卷积网络,极大减少了模型的运算量并保持了较高的识别精度,适合硬件资源有限的微型计算机.神经网络经过训练、调参、剪枝优化后,图像识别准确率达到91.9%,推理延时小于1.5 s.这种人工智能系统可以代替人工操作让顾客进行自主购物,适合在无人超市或者避免聚集接触的环境下使用.

This paper designed a deep convolutional neural network(DCNN)-VegNet to detect fruits and vegetable and implemented the network on a microprocess computer.VegNet consists of depthwise separable convolutions and inverted residual blocks,this architecture reduces the large amount of network parameters and fits microprocess computer due to the limitation of hardware resources.Experiments demonstrate AI technologies are able to be extended to the last inch equipment with robust and accurate performance.The detection rate of the AI based electronic scale to the trained targets reach 91.9%,and the recognition delay is less than 1.5 seconds.