Abstract

Abstract: Uilized PC vision and deep learning methods to choose and plant sound billets, which expanded plant populace and the yield per hectare of sugarcane planting. We utilized notable convolutional neural network (CNN) structures to deal with enormous picture datasets and move learning methods to extend the outcomes to various sugarcane assortments. It would be extremely tedious to gather and mark huge datasets for every sugarcane assortment, for which quality investigation is required, preceding planting. We utilized a two-venture move learning interaction to stretch out the prepared design to new assortments. We looked at results got during move learning utilizing AlexNet, VGG-16, GoogLeNet, ResNet101 structures to traditional PC vision techniques. Our objective was to decide the best way to deal with identify harmed and great billets in the most brief preparing time. Best brings about both time and exactness were gotten with AlexNet. For AlexNet, we looked at stages of three sugarcane assortments to track down the best model to distinguish the sound sugarcane billets. We at that point diminished the quantity of pictures utilized to retrain the model to decide tradeoff among time and execution. Eventually, one requirements a couple dozen billets of the new assortment to retrain the network. Our methodology prompted significant augmentations in the yield per hectare going from 33 to 80% contingent upon sugarcane assortmentKeywords: —Agricultural robotics, computer vision, convolutional neural networks, sugarcane, transfer learning