*4.4. Comparison of Detection Speed of the Models*

We compared the detection speed of the improved Yolov5s-CA and the original Yolov5s model (Table 4). The predicted inference speeds of the Yolov5s-CA model are 11.4 ms, 12.3 ms, and 11.9 ms, which are 2.3 ms, 1.6 ms, and 1.4 ms longer than the Yolov5s for the field-collected, synthetic, and mixed datasets, respectively. In addition, despite the increase in the parameters of the model, the size of Yolov5s-CA model is only 0.1 MB larger than the Yolov5s model. The detection speed or frames per second (FPS) of the Yolov5s-CA model was slightly lower, but had an improved performance compared to the Yolov5s model (Table 4). Therefore, the improved Yolov5s-CA model can ensure real-time performance with relatively little additional detection time and nearly no computational overhead.


**Table 4.** Performance comparison of model detection speed.
