Image recognition method using Local Binary Pattern and the Random forest classifier to count post larvae shrimp

Abstract

Thailand is one of the world's largest shrimp producers. However, Thailand's fishery industry, which includes the shrimp industry, is still using human labor intensively. The culture process starts with getting shrimp that are at the early “post-larvae” stage from a supplier and growing them in a closed environment. One of the most basic tasks, counting the number of larvae, is tedious labor-intensive work. To increase the effectiveness of this traditional task, the use of image recognition techniques for larvae counting was investigated through the evaluation of two feature extraction methods: Local Binary Pattern (LBP) and Red, Green, Blue (RGB) feature extraction, with an ensemble tree classifier, Random Forest (RF). Assessing the results with K-fold cross validation (k=5) showed the LBP method was 98.50% accurate. This research had a relatively large root mean square error (RMSE) of 14.43 due to the overlapping of shrimp. This method has the potential to help with this basic and essential task and increase the efficiency of the shrimp farming process.