Abstract
In Hiroshima Prefecture, lemon production is thriving. One key criterion for harvesting lemons is their size, measured by the diameter of the lemon’s cross-section. Currently, harvesting decisions are made by passing a metal ring with the minimum shipment standard diameter through the lemons to see if they pass through. There are two main problems associated with using this ring. The first problem is that the ring can scratch the surface of the lemons, leading to a decrease in quality. The second problem is that all lemons that cannot be visually assessed for harvesting must be passed through the ring, resulting in a high workload. This study utilizes RGB-D cameras to capture both RGB and depth images. By using Deep Neural Network (DNN) techniques, the lemon regions are detected from the images. Subsequently, regression methods are employed to estimate the diameter length as the lemon size from the extracted features of the lemons.
