Garbage Disposal of Complex Background Based on Deep Learning With Limited Hardware Resources

With the increasingly severe problems of garbage contamination, automatic garbage identification has significant value in actual application. However, the real implementation is still confronted with the limitations of equipment performance and the need for more computer resources. To effectively tackle the challenges, this paper proposes a model that uses the MobileNet as a base network to improve the performance of Mask R-CNN, which could be applied to the low-cost embedded devices. For a numerical evaluation on real images, we construct a garbage image data set containing multiple morphological types based on the complex background. Specifically, when pre-processing the data set, we use the methods of data expansion and transfer learning. Consequently, high-precision detection of garbage could be achieved under the condition of a few samples. The experimental results present that the approach can achieve a detection accuracy up to 93%. Furthermore, the proposed approach is a complete system combining both hardware and software design, which dramatically promotes the automation of production and provides a reference for the intelligent garbage sorting machines.