Research on ConvNeXt-L-CA Method for Image Classification Detection in Grain Silos

In recent years, in response to the country's requirements for innovating grain supervision methods and realizing "penetrating" supervision, the dynamic monitoring technology for grain storage quantities based on the video monitoring systems in the grain silos and computer vision has gradually been applied in the grain industry. In the practical application, it is found that there are many abnormalities in the warehouse images collected by the video surveillance system, and it is urgent to intelligently classify these warehouse images and maintain the video surveillance system to improve the accuracy of grain quantity monitoring in the warehouse. This paper used the ConvNeXt model as the backbone network, introduced the CA attention mechanism and Lion optimizer and proposes an improved image classification method for grain silos based on ConvNeXt. Experimental results show that the accuracy, precision, recall and F1 index of the improved ConvNeXt model reach 98.24%, 98.00%, 98.04% and 98.00%, respectively, which is 0.53% higher than that of the original ConvNeXt model, which verifies the effectiveness of the method and provides technical support for further enhancing the reliability and accuracy of grain information supervision.