Machine vision solutions for monitoring pest snails in Australian no-till cropping fields: An exploration of spectral characteristics and detectability

Abstract

In the realm of Australian agriculture, the invasive presence of two round snail species, Cernuella virgata and Theba pisana, and two conical snail species, Cochlicella acuta and Cochlicella barbara, has inflicted substantial economic losses. Effective integrated snail management hinges on precise monitoring of snail populations, yet current manual sampling techniques are prohibitively labour-intensive and susceptible to errors. This study underscores the imperative for a machine vision system capable of accurately and efficiently observing snail populations in the diverse landscapes of Australian broadacre no-till cropping fields. The inherent challenges of varied plant residues, gravels, and soil types in crop fields demands a sophisticated optical system to detect small, clustered, and camouflaged snails. Through a meticulous exploration of spectral features and detectability, employing different combinations of wavelengths and machine learning algorithms, this study yielded the promising findings that live round snails, dead round snails, live conical snails, dead conical snails, and complex field materials, were classified with F1 scores ranging from 0.7 to 0.9. This study not only highlights the potential of such a machine vision system, but also delineates ongoing challenges that warrant further investigation. The insights derived serve as a guide for the development of a robust machine vision system aimed at mitigating the impact of pest snails in agricultural fields.

Keywords

• Snail detection
• Integrated pest management
• Hyperspectral imaging
• Machine vision
• Machine learning