Efficient Patch-Wise Crop Detection Algorithm for UAV-Generated Orthomosaic

Effective image segmentation is essential for extracting semantic information and performing comprehensive analytics on the collected data. This chapter presents novel patch-wise segmentation techniques for efficient crop detection on high-resolution remote sensing farm orthomosaic images. The proposed algorithms slice the farm orthomosaic with varying overlaps and employ the Mask R-CNN segmentation to identify and detect crops. The first algorithm slices the farm orthomosaic with no overlap, resulting in the loss of data at the slicing boundaries. The second algorithm slices the farm orthomosaic with an overlap equal to the width of the largest crop, which is determined by randomly selecting n% (variable between 0.1 and 0.9) of the available slices. The third method takes into account the entire farm in order to determine the width of the largest crop and hence the overlap. The results reveal that Algorithm 3, which uses the width of the largest crop on the farm as the overlap, produces the best possible segmentation. For values of n greater than 0.6, the performance and quality of the resulting mask generated by Algorithms 2 and 3 are comparable. Algorithm 1 has significantly worse performance than Algorithms 2 and 3. This research offers a valuable framework for efficient crop detection on high-resolution remote sensing farm orthomosaic images, which may be used to predict farm yield, monitor crop health, and map weed cover areas.