To address the problems of low recognition accuracy,large number of model parameters,slow inference speed and dif-ficult deployment in current convolutional neural network seed sorting methods,a seed sorting method based on lightweight pyramidal dilated convolutional network is proposed.A residual spatial pyramid module is proposed to expand the perceptual field by using the convolution of dilated with different expansion rates,to effectively extract the multi-scale features.Then,deep-wise separable convo-lution techniques are used to reduce the model parameters and the computational complexity.A lightweight attention mechanism mod-ule is introduced into the network structure to improve the extraction of seed key feature,the local cross-channel interactions are a-dopted to focus on the important information.The experimental results show that the parameter quantity of the proposed network is only 0.13 M,with a accuracy on corn dataset and red kidney bean dataset of 96.00％and 97.38％,and the average time of 4.51 ms to recognize single image on NVIDIA Quadro board,the recognition time on the NVIDIA Quadro board is better than that of the ma-instream lightweight networks,such as MobileNetv2,Shufflenetv2 and PPLC-Net,etc.,which can meet the requirements of real-time recognition in industrial sites.
维普
万方数据