Precise Real-Time Detection of Nonforested Areas With UAVs

摘要

This paper presents a new method for real-time automatic detection of nonforested and eroded areas in tropical rain forests. It is based on simple image algebra between color components, which enhances the contrast between brown and green areas. A successive segmentation through multiple thresholds, based on newly proposed indices of “brown color excess” and the “nonforest detection index,” leads to a binary map that clearly identifies forested and nonforested areas. Experimental tests, performed and compared with other recommended methods based on: region growing, active contours, and clustering, outperformed in detection accuracy (Fm = 96.4%) and processing times (Tp = 0.082 s). The method presented copes well with detecting regional irregularities and reduces frequent issues of nondetection, as well as false positives caused by intensity changes, shadows, and/or partial occlusions. The low processing times achieved with the proposed method allow real-time applications for low-cost unmanned aerial vehicle and unmanned aircraft systems with conventional camera equipment.