Line Detection and Segmentation of Annual Crops Using Hybrid Method

Currently, agriculture has incorporated more and more technologies to optimize production, reducing waste and environmental impacts of this activity. Precision agriculture aims to maximize efficiency and agricultural productivity while reducing the use of chemicals. This work proposes an automatic approach based on supervised learning and image processing methods for crop line detection in frontal images obtained by a ground camera. Along with this, data augmentation techniques are applied to increase the reference method from the literature ability of the algorithm. The proposed method uses a semantic segmentation architecture based on Deeplabv3+ architecture to segment the region of interest, and the Canny Edge technique is applied to locate the image edges. Finally, the probabilistic Hough transform is applied to identify the segments from the identified boundaries. The proposed method is evaluated and compared with a variation of a reference method from the literature algorithm, where our method presented higher performance, achieving an overall accuracy of 84.1%, even on images with high weed presence and crop line gaps.