A coarse-to-fine leaf detection approach based on leaf skeleton identification and joint segmentation

Plant leaf detection and segmentation are challenging tasks for in-situ plant image analysis. Here, a novel scheme is proposed to detect individual leaves accurately determine shapes in natural scenes. A skeleton-extraction method was developed by analysing local features of skeleton pixels. Approximate positions were determined according the main skeleton. Sub-images containing only single target extracted from whole images position size Accurate analysis conducted on sub-images leaves. Leaf direction calculated examining structure Joint combining region active shape model presented elucidate shape. implemented using deep learning approach, Faster R–CNN. dataset four types different complexity built evaluate algorithms. with occlusions complex backgrounds effectively detected their determined. Detection accuracy 81.10%–100%, 86.75%–100% The demonstrated comparable ability that Furthermore, rates success our ranged between 89.06% 100%, while average measurement difference 1.29° compared manual measurement. 75.95%–100% all images. Therefore, this accurate stable precise measurements agricultural applications.