2.2. 4 Applications of Deep Learning Based Semantic Segmentation of Images

Abstract

Deep convolutional neural network is demonstrated on two problems: semantic segmentation of agricultural images for weed detection and semantic segmentation of garbage in images.  Weed segmentation is important since it allows detection of weed infestation in agricultural plantations and enables farmers to perform targeted herbicide application.  Garbage detection is important to create applications that would allow easier reporting of littered sites to the authorities and increase the public awareness about the problem.  Using transfer learning methods improved the model accuracy for weed segmentation, and showed great potential for application of this method using cheap sensors on farms.  The algorithm for garbage detection achieved high accuracy for classification of different garbage types, allowing the potential deployment of this system on cloud network.