Deep Learning and Remote Sensing: Detection of Dumping Waste Using UAV

An important success and use of Deep Learning in recent years has been in the field of image processing. Research on Deep Learning has shown that these architectures particularly convolution neuron network (CNN) can learn solutions with human-level capability for certain visual tasks. These techniques have been used in particular in remote sensing image analysis tasks, including object detection on images, image fusion, image recording, scene classification, segmentation, object-based image analysis, land use and land cover classification (LULC). In this paper we present an automatic solution for the detection of clandestine waste dumps using unmanned aerial vehicle (UAV) images in the Saint Louis area of Senegal, West Africa. This is a challenging task given the very high spatial resolution of UAV images (on the order of a few centimeters) and the extremely high level of detail, which require suitable automatic analysis methods. Our proposed method begins by 1) segmenting image into four (4) regions, which can be used as an input image 2) Reduce size of input images into 300x300x3 for the CNN entries 3) Labelling the image by determining region of interest. Next Single shot detector SSD is used to mine highly descriptive features from these datasets. The results show that the model recognizes well the areas concerned but presents difficulties on some areas lacking clear ground truths. (c) 2021 The Authors. Published by Elsevier B.V.

Automated summary

Deep learning, especially convolutional neural networks (CNN), has been successfully applied in image processing, particularly in remote sensing image analysis tasks. This paper presents an automatic solution for detecting clandestine waste dumps using UAV images and utilizes CNN and SSD to achieve good recognition results.