Computer Vision and Spectral Analysis Technologies for Non-Invasive Plant Studying

Abstract

Computer vision and spectral analysis of digital images are technologies that allow the use of automated and robotic systems for non-invasive plant studying, production and harvesting of agricultural products, phenotyping and selection of new plant species. (Research purpose) The research purpose is in analyzing the application of modern digital non-invasive methods of plant research using computer (technical) vision and prospects for their implementation. (Materials and methods) Authors have reviewed the works on the use of non-invasive methods for obtaining information about the state of plants. The article presents classification and analyze of the collected materials according to the criteria for collecting and analyzing digital data, the scope of application and prospects for implementation. Authors used the methods of a systematic approach to the research problem. (Results and discussion) The article presents the main directions of using computer vision systems and digital image analysis. The use of computer vision technologies in plant phenotyping and selection reduces the labor cost of research, allowing the formation of digital databases with a clear structure and classification by morphological features. It was found that the introduction of neural networks in the process of digital image processing increases the accuracy of plant recognition up to 99.9 percent, and infectious diseases up to 80 percent on average. (Conclusions) The article shows that in studies using hyperspectral optical cameras and sensors are used cameras with an optical range from 400 to 1000 nanometers, and in rare cases, hyperspectral camera systems with a total coverage of the optical range from 350 to 2000 nanometers. These optical systems are mainly installed on unmanned aerial vehicles to determine vegetation indices, foci of infection and the fertility of agricultural fields. It was found that computer vision systems with hyperspectral cameras could be used in conjunction with fluorescent plant markers, which makes it possible to solve complex problems of crop recognition without involving computational resources.