Shrimp-growth Estimation Based on ResNeXt for an Automatic Feeding-tray Lifting System Used in Shrimp Farming

Abstract

The shrimp agriculturists monitor shrimp growth by observing the size of shrimps in the feeding tray with the naked eye. This approach is time-consuming and needs experienced workers. This study proposes an automatic approach for estimating shrimp size using images. A mask region-based convolutional neural network with ResNeXt was trained to detect shrimps in an image. The detection model achieved an overall precision of 74.45%, recall of 72.20%, Fl score of 73.31 %, and AP of 69.04%. The two unique methods were proposed for estimating shrimp size. The first method achieved a mean absolute error of 0.30 cm and a mean absolute percentage error of 3.97%. The second method achieved a mean absolute error of 0.35 cm and a mean absolute percentage error of 4.59%. The proposed system achieved an automatic shrimp size estimation from the image and provided helpful information for agriculturists.