Abstract

Litchi sorting image recognition and classification plays a vital role in agricultural production. First, a dataset of lychee images containing different shapes, colors and blemishes was collected and constructed. Then, through data preprocessing, the image is scaled, clipped and normalized to improve the robustness of the model. The convolutional layer, pooling layer, and fully connected layer were used to construct deep convolutional neural networks to recognize complex features of litchi images. Throughout the model's training phase, the backpropagation algorithm is employed and subsequently assessed using the test set, resulting in commendable levels of accuracy and precision. Finally, the trained model is successfully deployed to the actual litchi sorting system, and the automatic recognition and classification of different kinds of litchi is realized. This study provides an efficient and accurate litchi sorting solution for agricultural production, which has a wide application prospect.