Noise reduction and accuracy improvement in optical-penetration-based silkworm gender identification

Abstract

This work proposes and experimentally demonstrates a very high-accuracy optical-penetration-based silkworm gender identification method. The key to our approach lies in our optical noise suppression scheme that involves a combination of optical image magnification, polarization filtering, and automatic selection of the optical region of interest (ROI). In the experimental demonstration, an optical image magnification of 5.65 and polarization filtering are achieved by merely using a 10-mm focal length lens and crossed polarizers, respectively. Further, the optical ROI is automatically generated under white light illumination. The ROI is also fitted to the size of the body of the silkworm pupa, thereby assisting in removing unwanted image noise when the silkworm gender analysis is performed under red light illumination. In addition, to perform silkworm gender identification, simple image processing operations including color image thresholding, an overlaying process of the optical ROI, local image thresholding, blob filtering, and centroid analysis are employed. Our experimental results show a very promising accuracy of 97.9% in identifying the silkworm gender with a fast processing time of 54.9 ms. Other important features of our method include low cost, ease of implementation, and simplicity in terms of process control.