Grape Bunch Detection Using A Pixel-Wise Classification In Image Processing

This work presents a technique for grape bunches detection within Images. The approach is based on a pixel-wise classification boosted with a morphology operator. Color indices, commonly used for plant segmentation in image processing are proposed here for separating grape pixels from background, along with color components from color spaces. These color features are the input for the classification process. The proposed pipeline achieves an accuracy of 0.901 on images similar to the ones used for training, and 0.84 on images from a different dataset. Finally, we use an area filtering for noise-handling, outputting the grape bunches localization.