The combination of two YOLOs and ANN resulted in more successful object detection than relying solely on one YOLO, even in cases where only one YOLO struggled to detect an object or made incorrect detections.
Deep learning-based object detection has numerous practical applications and continues to be a subject of ongoing research in the pursuit of improved performance. However, it is not always possible to detect objects as a whole due to the training process, object differences, obstacles, etc. For this reason, the detection success automatically decreases. This study proposes a method using two YOLOs to detect entire objects and their distinctive regions separately to increase detection success. One YOLO’s target is wider and more accurate, while the other detects a smaller and more difficult area of objects. An ANN evaluates the accuracy of the results because the YOLOs have different prediction scores. As experimental results, the combination of two YOLOs and ANN resulted in more successful object detection than relying solely on one YOLO, even in cases where only one YOLO struggled to detect an object or made incorrect detections. In other words, not only correct object detection was achieved, but also false detections were prevented. In addition, this method is a method that can be used without extra software knowledge.