Automatic shrimp shape grading using evolution constructed features

Shrimp has become one of the most favorite seafood in recent years. As worldwide shrimp production grows, shrimp quality evaluation becomes a critical task for the seafood and aquaculture industries. Automatic evaluation of shrimp shape is critical to improving shrimp quality and production efficiency. This paper proposes an Evolution COnstructed (ECO) features based method to automatically evaluate shrimp shape completeness. Rather than depends on human expert-designed features or deliberated image processing techniques, the proposed method automatically constructs features that are used by AdaBoost model to detect broken shrimp. Experimental results show that ECO features based method obtains a 95.1% overall classification accuracy with a 0.948 precision rate and a 0.920 recall on the 879 shrimp samples collected for testing. Although the experiment was performed on one select species to prove feasibility, the proposed method can be easily adapted to other species.