| Abstract: | Real-time detection and identification of invertebrates on crops is a useful capability for integrated pest management, however, this challenging task has not been solved. Compared with other technologies, a machine vision system (MVS) could provide a more flexible solution. To date, most studies have focused on counting and identifying specimens in sample containers, glass slides or traps where the illumination and background reflection can be well controlled; few studies have been conducted to detect pests on plants. In the context of invertebrate detection or identification, the spectra of visible light, near infrared (NIR) and soft X-ray have been well studied, while the spectrum of ultraviolet (UV) is still untouched. Many species of bird prey on invertebrate pests and have adaptations in their visual system to enhance detection of targets. These birds can use both UV and visible light to hunt. If the mechanisms of bird vision could be transferred to a technological visual system, it might improve the capability for invertebrate detection. This study provides an initial estimation of the contribution of UV for invertebrate detection on green leaves. By fusing the UV images into the visible light and NIR images, the MVS can detect nine invertebrate species on leaves of plants and the UV images can significantly reduce segmentation errors. The initial experiment was conducted in a laboratory, however, this study shows promise for infield applications. |
