YOLOAPPLE: Augment Yolov3 deep learning algorithm for apple fruit quality detection

Apple detection helps the food manufacturing process to distinguish between fresh and damaged apples. In this modern world, many apple-detecting flaws are discovered before harvest. After a harvest, the system is required to identify apple species and quality, which will help in food production machinery. In this research, a novel YOLOAPPLE has been proposed for identifying different apple objects such as three classes: normal apple, damaged, and red delicious apple using Augment Yolov3. Using Grab cut to remove the background of the apple for better results in the next iteration. The augment Yolov3 with extra spatial pyramid information and a swish activation function to maintain feature loss preferences throughout training. Yolov3 is improved by the Darknet53 convolution neural network acting as a backbone and by adding spatial pyramid pooling features using the feature pyramid network before the object detector. Finally, the fully connected layer will classify as normal apple, damaged, and red delicious. The Augment Yolov3 model enables multi-class detection and recognition system that achieves higher mean average precision of 99.13% when compared to the conventional Yolov3, Yolov4 deep learning model. The experimental results originated from a newly generated object recognition model that was created by utilizing the Kaggle dataset using Google Colab inference on an NVIDIA Tesla K-80 GPU for a better localization process and its precise multi-object detection.