The Waste Detection System of Shrimp Feeding with a Waterproof Camera using Yolo Algorithm

Abstract

This study aims to build a waste detection system of shrimp feeding using the Yolo algorithm. Shrimp are not efficient in utilizing their feed, which is only 70-80%. The remaining shrimp feed will become waste that causes decay and decreased water quality due to the accumulation of high organic matter and toxic compounds, namely nitrite (NO2) and ammonia (NH3). Therefore, a feed waste detection system was developed in this study using You Look Only Once (YOLO). The initial stage is to collect images in the water as training data with a waterproof camera. Furthermore, the image is marked with a Yolo mark. Then training is carried out where it is extracted with the Convolutional neural network layer, which is used as input into the Fully Connected Layer. The output is a weight file that will be used to detect shrimp feed. From the tests carried out, the system produces 96-97% mean average precision (mAP) values at max batches of 4000-10000. Another result shows that the best mAP was obtained at a distance of 25 cm with mAP value of 82.31%.