Please use this identifier to cite or link to this item:
|Scopus||Web of Science®||Altmetric|
|Title:||A multispectral machine vision system for invertebrate detection on green leaves|
|Citation:||Computers and Electronics in Agriculture, 2018; 150:279-288|
|Huajian Liu, Javaan Singh Chahl|
|Abstract:||Detection and identification of invertebrate pests in farming fields is a prerequisite necessity for integrated pest management (IPM), however, current sensing technologies do not meet the requirements for IPM. Currently, farmers have to first sample pests and then manually count and identify them, in a way that is time-consuming, labour-intensive and error-prone. Machine vision technology has taken over part of the work in a more efficient and accurate manner. However, current machine vision systems (MVSs) have limitations in detecting pests on crops and the counting and identification are constrained in laboratories or pest traps, resulting in the exact time and locations of pests being unknown, hindering more proper decisions and efficient actions. In this study, we developed a multispectral MVS to detect common invertebrate pests on green leaves in natural environment. First, it was found that, besides visible light and near-infrared, the ultraviolet is a good indicator to distinguish green leaves from other materials. Then for multispectral or hyperspectral data processing, we proposed two models, one named normalised hypercube and another named hyper-hue, which are less affected by uneven illumination and can reflect data distribution, resulting in more accurate classification than the normal method of spectral angle mapper (SAM). Further, the relationship between spectral angle and the relative angle of hyper-hue was studied and it was found that usually, data of hyper-hue has larger inter-class distances which could contribute to better classification. At last, to solve the practical problems of image registration and real-time infield applications, instead of registering 2D images, the MVS created and registered 3D point clouds. In an experiment of detecting twelve types of common invertebrate pests on crops, the proposed MVS showed acceptable accuracy.|
|Keywords:||Machine vision; computer vision; multispectral imaging; 3D vision; invertebrate detection; insect detection Integrated pest management; precision agriculture|
|Rights:||© 2018 Elsevier B.V. All rights reserved.|
|Appears in Collections:||Agriculture, Food and Wine publications|
Files in This Item:
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.