A lightweighted method based on the optimized MobileViT model was proposed to solve the challenges in identifying weeds from crop seedlings in agricultural environments.SimAM attention mechanism was introduced to enhance the model's ability to pay attention to features.SCConv convolution module was used to reduce the spatial and channel redundancy of features in convolutional neural networks to lower computational costs and model storage， while improving the performance of the convolution module.A loss function strategy combining Label Smoothing Loss and Cross Entropy Loss was proposed to improve the generalization performance of the model， reduce the risk of overfitting， and accelerate the convergence process of the model.Images of 12 common crop seedlings and weeds in the field were used as the training dataset to evaluate the performance of the improved model MobileViT-SS.The results showed that the average recognition accuracy， precision， recall rate， and the F₁ score of the improved model reached 95.91%， 95.97%， 95.46%， and 95.69%， respectively， all of which were superior to that in the widely used deep neural network models including VGG-16， ResNet-18， and MobileNetv3.It is indicated that the improved model MobileViT-SS can accurately and quickly distinguish various weeds from crop seedlings with similar morphology.It will provide technical reference for the identification of weeds from crop seedlings with similar morphology.