Computer Vision Drives the New Quality Productive Forces in Agriculture: A Method for Recognizing Farming Behavior on Edge Computing Devices

In the context of rapid advancements in new quality productive forces in agriculture, precise management of the crop growth cycle is crucial for enhancing crop yields and the economic benefits of plantations. However, in the Southwest region of China, agricultural enterprises primarily rely on manual record-keeping to manage farm labor. This approach is inefficient and cannot guarantee data accuracy, severely limiting the improvement of agricultural productivity. In response to this problem, this paper proposes an intelligent solution based on deep learning. We developed a novel network called Slowfast-a, which integrates the Slowfast behavior recognition algorithm with attention mechanisms and optimizes the model size through knowledge distillation technology. This enables deployment on edge computing devices in the field. Our solution not only overcomes the limitations of traditional Convolutional Neural Networks (CNNs) in recognizing blurred views but also addresses the computational resource constraints of edge devices. Experimental results demonstrate that on edge computing devices, this model achieves a farming behavior recognition accuracy of 93.50%, with an average recognition time of only 0.25 s. This study significantly enhances the level of intelligence in agricultural production operations, meets the demand for efficient and precise management in the new era of agricultural productivity, and provides efficient and practical technical support for modern agricultural production. This, in turn, has the potential to drive further development in new quality productive forces within agriculture.