Single Plant Fertilization using a Robotic Platform in an Organic Cropping Environment

Summary

The increasing demand of organic vegetables has driven conventional farmers to change their businesses in order to develop organic methods for cultivation and the use of alternative techniques to avoid damaging the soil and the quality of the products. In some cases, to enhance biodiversity and soil fertility, crops are established in a mixed pattern called ‘strip cropping’ where single or dual lines of a given species are alternated with a second compatible species, with the aim to enhance resilience, system sustainability, local nutrient recycling, and soil carbon storage. However, this husbandry of crops grown and mixed in a strip design pose new challenges regarding mechanisation, which in many cases can only be overcome by increasing human tasks. To counteract the additional labour of a multi-crop system, one of the main objectives of the ‘Sureveg’ CORE Organic Cofund ERA-Net European project is to evaluate the benefits of growing in alternate rows for the production of organic vegetables and includes the use of robots as a tool to facilitate the automation of the process, allowing the individual treatment of organic fertilization at plant level. Within the project framework a modular proof-of-concept version has been produced, combining several sensing technologies (3x LiDAR, plus a multispectral RGB-NIR camera) with actuation in the form of a robotic arm operating upside down. The present work describes a method to develop fertilization tasks with recycled organic waste in strip-cropping farms, based on detection of plant species (cabbage) using and liquid fertilizer application with a robot. For that, it is necessary to identify the crop (variable in the rows) and place the robot with respect to the plants to apply the product. Test fields were located at ETSIAAB - UPM (40°26’33.1"N, 3°43’41.9"W) where stripped crops were established and monitored along the growing season. In order to detect autonomously each single plant, point clouds of the three LiDAR units were combined, soil was removed applying a weighted threshold, and plants were identified using clustering and convolutional neural network methodologies. To trigger the actuation system, the decision on which plant had to be sprayed with the liquid fertilizer was taken according to two factors: 1) the estimated volume of every single plant, and 2) the multispectral indexes calculated using the RGB-NIR camera. The prototype is fully functional, and further test are needed to quantify its performance