This paper presents a novel, low-cost, user-friendly Precision Agriculture platform that attempts to alleviate the drawbacks of limited battery life by carefully designing missions tailored to each field’s specific, time-changing characteristics. The proposed system is capable of designing coverage missions for any type of UAV, integrating field characteristics into the resulting trajectory, such as irregular field shape and obstacles. The collected images are automatically processed to create detailed orthomosaics of the field and extract the corresponding vegetation indices. A novel mechanism is then introduced that automatically extracts possible problematic areas of the field and subsequently designs a follow-up UAV mission to acquire extra information on these regions. The toolchain is finished by using a deep learning module that was made just for finding weeds in the close-examination flight. For the development of such a deep-learning module, a new weed dataset from the UAV’s perspective, which is publicly available for download, was collected and annotated. All the above functionalities are enclosed in an open-source, end-to-end platform, named Cognitional Operations of micro Flying vehicles (CoFly). The effectiveness of the proposed system was tested and validated with extensive experimentation in agricultural fields with cotton in Larissa, Greece during two different crop sessions.

No abstract