MAIZE SEEDLING COUNTING ON UAV RGB IMAGERY USING COMPUTER VISION AND DEEP LEARNING UNDER SUBSTANTIAL WEED INFESTATION AND PARTIAL OCCLUSION
I. A. Pestunov1,2, R. A. Kalashnikov1, R. A. Mukhamediev3,4, A. Symagulov3,4
1Federal Research Center for Information and Computational Technologies, Novosibirsk, Russia
2Institute of Automation and Electrometry, Siberian Branch, Russian Academy of Sciences, Novosibirsk, Russia
3Institute of Information and Computational Technologies CS MSHE RK, Almaty, Kazakhstan
4Satbayev University, Almaty, Kazakhstan
Keywords: RGB images, UAV, maize seedlings counting, semantic segmentation, skeletonization, graph features, DeepLabV3+, Random Forest, SVM
Abstract
An automatic method is proposed for counting maize seedlings under conditions of substantial weed infestation and partial occlusion using ultra-high-resolution (<0.5 cm/pixel) RGB imagery acquired from an unmanned aerial vehicle. The method is based on a combination of computer vision and machine learning algorithms. Experimental results demonstrate that the accuracy of estimating the number of maize seedlings at early growth stages averaged 97%.
|
