Aplikace pro detekci rostlin využívající metody a nástroje umělé inteligence

Abstract

This thesis focuses on detecting plants from image data using artificial intelligence methods and tools. As a part of the work, dataset was available, which was preprocessed - this included resizing all images to a uniform resolution of 1024 x 1024 pixels and removing distracting artifacts, particularly black marker labels used to mark Petri dishes. Three models of YOLO architecture were trained. A comparison of the individual models was done. After that followed calculation of plant size using the height of bounding boxes and also using segmentation masks, which were intended to provide more accurate size estimation. For the segmentation masks, the SAM2 model used. After that the masks were skeletonized and plant lenght was estimated by BFS algorithm.