A Classification Algorithm of Grain Crop Image Based on Improved SqueezeNet Model

Aiming at the disadvantages of traditional crop screening methods, such as dependence on experience and low accuracy, a granular crop classification algorithm based on deep learning was proposed. This paper took peanuts as an example to construct an image data set. According to the damage of the peanut skin, it could be divided into two types: intact peanuts and peanuts with damaged skin. Then based on the Fire module in SqueezeNet, the convolutional neural network was improved. Using comprehensive optimization training strategies such as Msra initialization, RMSprop method and momentum method, the classification accuracy rate could reach 94.4%. Compared with the existing crop screening methods, the accuracy rate was significantly improved, and it was helpful to realize the screening and identification of granular crops.