The classification of weeds and crops is key to precision agriculture. To achieve the challenge of recognizing and classifying weeds and crops with high accuracy, this paper uses a deep learning approach. We propose the SE-ResNext network to optimize ResNext using the channel attention mechanism. Embedding the channel attention mechanism in the backbone network of ResNext, this deep learning model achieves 97.51% classification accuracy on the V2 Plant Seedlings dataset. The training accuracy of the SE-ResNext model is 98.49% and the MSE of the SE-ResNext model is 0.044, which are the best values. Compared to AlexNet and GoogleNet, this model has higher classification accuracy. The results can provide technical support for the autonomous operation of field-weeding robots. Enables accurate classification of weeds and crops.