Comparison of Different Deep Learning Object Detection Algorithms on Fruit Drying Characterization

Abstract

Object detection is an essential task in the field of computer vision and a prominent area of research. In the past, the categorization of raw and dry Tamanu fruits was dependent on human perception. Nevertheless, due to the progress in object detection, this task can currently be computerized. This study employs three deep learning object detection models: You Only Look Once v5m (YOLOv5m), Single Shot Detector (SSD) MobileNet and EfficientDet. The models were trained using images of Tamanu fruits in their raw and dry state, which were directly collected from the dryer device. Following the completion of training, the models underwent evaluation to identify the one with the highest level of accuracy. YOLOv5m demonstrated superior performance compared to SSD MobileNet and EfficientDet, achieving a mean average precision (mAP) of 0.99589. SSD MobileNet demonstrated exceptional performance in real-time object detection, accurately detecting the majority of objects with a high level of confidence. This study showcases the efficacy of employing deep learning object detection models to automate the classification of Tamanu fruit.