The advancement of pest detection in smart agriculture has outpaced the development of technologies for detecting avian crop threats. Recently, the rise in quelea populations due to ecological improvements has resulted in significant agricultural damage as these birds feed on grains and seedlings of crops such as millet, maize, and rice. This study presents an artificial intelligence-based quelea recognition method suitable for embedded systems. This research enhances the Tiny-YOLOV3 network, developing the Enhanced Tiny-YOLO (ET-YOLO) for real-time detection of queleas in complex outdoor environments. A dataset comprising 3, 500 high-resolution quelea images, taken under various conditions and distances, was used for training and testing. Experimental results demonstrate that ET-YOLO achieves an average detection accuracy of 88. 5% and a speed of 62 frames per second in video feeds, improving accuracy by 15 percentage points and speed by 2 frames per second over Tiny-YOLOV3. Additionally, ET-YOLO outperforms SSDMobileNetV2, YOLOV3, and Faster-RCNN, with accuracy improvements of 17, 1. 6, and 1. 4 percentage points, and speed increases of 1, 35, and 44 frames per second, respectively. With a model size of 56 MB, ET-YOLO is well-suited for deployment in embedded systems within agricultural robots and smart machinery. The comparison shows that the ET-YOLO network proposed in this paper achieves higher detection accuracy and speed than the original Tiny-YOLOV3 lightweight target detection network.
Onu et al. (Thu,) studied this question.