Field robot to detect plants infected by Candidatus Liberibacter solanacearum in horticultural crops using multispectral computer vision
Author
Cubero, Sergio; López, Santiago; Marco-Noales, Ester; Alegre, Vicente; Barbé, Silvia; Aguilar, Enrique; Navarro-Herrero, Inmaculada; Aleixos, Nuria; Blasco, JoséDate
2018Cita bibliográfica
Cubero, S., López-Alamán, S., Marco-Noales, E., Sanjuan, S., Alegre, V. Barbé, S. et al. (2018). Field robot to detect plants infected by Candidatus Liberibacter solanacearum in horticultural crops using multispectral computer vision. Proceedings of the European Conference on Agricultural Engineering (AgEng2018), 255-258.Abstract
A low cost field robot has been built capable of transporting a series of sensors through horticultural fields, with the aim of early detecting problems in the crop by means of proximal sensing techniques. The robot is operated by remote control and is driven by two electric motors coupled to the wheels and powered by batteries. The sensors include different thermal, colour and multispectral cameras in the visible and the near-infrared range that are synchronised with the advance of the robot by means of an encoder coupled to the axis of the motors. The position of each image is geolocated using a GPS. An industrial computer receives the encoder pulses and triggers the cameras, also receiving and storing the images and GPS information for further processing. The inspection area is located beneath the robot with the cameras focusing downwards (to the crop). To avoid the negative influence of direct sunlight, the area had been covered with a canvas and illuminated artificially with four-spot halogen lights. A telescopic extension system between 100 and 200 cm allows the robot to adapt to crops with different row widths. The first trials were carried out in a carrot test field located in Villena (Spain) to detect plants infected with Candidatus Liberibacter solanacearum. The crop was inspected every month from sow to harvest. Labels were placed on 100 plants to guarantee their individual identification in the images. During the harvest, these plants were collected separately, identified and analysed in the laboratory using molecular techniques in order to determine whereas they were infected or not. Several maps of the field have been created using spectral indexes at a very high resolutions between 0.5 mm/pixel and 2.5 mm/pixel depending on the camera.