Transfer Learning based Data-Efficient Machine Learning Enabled Classification

Recently, waste sorting has become more and more important in our daily life. It plays an essential role in the big picture of waste recycling, reducing environmental pollution significantly. Deep learning (DL) methods have been dominating the field of image classification and have been successfully applied to waste sorting tasks to achieve state-of-art performance. However, most traditional DL methods require a massive amount of annotated data for the training phase. Unfortunately, there is only one small data set for waste sorting, TrashNet created by Standford. In addition, manually collecting and labeling a massive data-set can be too costly. To address this issue, we decided to implement transfer learning (TL) techniques to construct a robust model based on a fairly small set of training data by transferring knowledge from existing deep networks, such as AlexNet, Resnet, and DensNet. As an innovation, we propose a novel domain loss function, Dual Dynamic Domain Distance (4D), to produce a more accurate domain distance measurement. There are three contributions to this paper. First, our model has achieved the best performance on the TrashNet data. Secondly, it is the first time that TL has been used for waste sorting. Finally, the proposed novel 4D domain loss has improved the performance of TL for this task. In this paper, we implemented two types of transfer learning methods, DDC, DeepCoral, to TrashNet data-set. Moreover, the DeepCoral-Resnet50 model yields the best performance of 96% test accuracy. More importantly, this work can be easily generalized to other image classification tasks.