Garbage localization based on weakly supervised learning in Deep Convolutional Neural Network
Mohd Anjum, M. Sarosh Umar · 2018 International Conference on Advances in Computing, Communication Control and Networking (ICACCCN) · 2018
Nowadays, Illegal dumping of garbage in residential area is a troublesome problem for both inhabitants and civic administration. So, government authorities are struggling to solve this issue in efficient way. The automatization of garbage collection process is an essential requirement and one of the major needs to proceed toward smart cities. The atomization of garbage collection requires the detection and localization of garbage automatically. The stated task is not an easy work as to train any machine learning system; it is needed corresponding dataset with its ground truth. The ground truth acquisition is a tedious task as it is needed to annotate every pixel in the image with its corresponding class. This needs a lot of manual work which is expensive in term of time and money both. In this paper, a garbage detection and localization system is proposed based on Convolutional Neural Network, which is trained on images labeled as garbage or non-garbage. The labeling of images in two classes as garbage or non-garbage is a simple and relatively fast task. The system yields better results in comparison to existing weekly-supervised methods; it also produces competitive results with respect to supervised learning method. The performance of the system is evaluated with pixel level of accuracy and it is also evaluated qualitatively by human experts.