As an important food and industrial raw material, cassava is widely cultivated worldwide. However, cassava plants are often attacked by various pests and diseases, resulting in significant yield losses. Therefore, timely identification of cassava plant pests and diseases is crucial. The accuracy of existing cassava pest detection technology needs improvement, especially when farmers capture images of cassava plants in different lighting and equipment conditions. These factors can affect the quality of the image thereby affecting the recognition results. To address this issue, a new cassava pest recognition system based on the improved YOLO-V5 algorithm is developed. By incorporating the attention mechanism module SE into YOLO-V5, the system's focus on the characteristics of the leaves has been enhanced, making it easier to detect pests and diseases. Moreover, the system is suitable for real-time environments and is easy to implement and use, making it possible to deploy it in various terminals and networked devices. Our experimental results demonstrate that the system performs well in detecting cassava pests and diseases.