Study of Underwater Fruit Object Detection Using Deep Learning Model

Abstract

Underwater autonomous vehicle operations are becoming progressively important in order to avoid the hazardous high-pressure deep-sea environment, and the relevance of underwater study and utilisation of marine resource is also rising. Computer vision is noteworthy technology for underwater autonomous vehicles study. In this research work, underwater raw data set is used for training, validating, and testing using YOLO v5 deep learning model to detect the one class (fruit) object. As the underwater images are blurry and hazy, detecting underwater objects without pre-processing is very challenging. In this study, we utilised raw data as underwater dataset to train the yolo model. The raw underwater dataset is difficult to acquire, so in the laboratory Raspberry-pi camera is used to capture the object at different angles, thereafter, data is augmented, yolo model is trained and performance parameters such as accuracy, precision, sensitivity and F1 score are analysed.