Automated Apple Recognition System Using Semantic Segmentation Networks with Group and Shuffle Operators

Apples are one of the most consumed fruits, and they require efficient harvesting procedures to remains in optimal states for a longer period, especially during transportation. Therefore, automation has been adopted by many orchard operators help process, which includes apple localization on trees. The de facto sensor that is currently used this task standard camera, can capture wide view information various trees from reasonable distance. paper aims produce output mask locations tree automatically using deep semantic segmentation network. network must be robust enough overcome all challenges shadow, surrounding illumination, size variations, occlusion accurate pixel-wise apples. A high-resolution architecture embedded with an optimized design group shuffle (GSO) best GSO allows reduce dependency few sets dominant convolutional filters forcing each smaller contribute effectively extracting features. experimental results show proposed network, GSHR-Net, two convolution applied layers produced mean intersection over union 0.8045. performance benchmarked 11 other state-of-the-art networks. For future work, increased integrating synthetic augmented data further optimize training phase. Moreover, spatial channel-based attention mechanisms also explored emphasizing some strategic apples, makes recognition more accurate.