Leveraging OpenStreetMap and Multimodal Remote Sensing Data with Joint Deep Learning for Wastewater Treatment Plants Detection

Abstract

Humans rely on clean water for their health, well-being, and various socio-economic activities. During the past few years, the COVID-19 pandemic has been a constant reminder of about the importance of hygiene and sanitation for public health. The most common approach to securing clean water supplies for this purpose is via wastewater treatment. To date, an effective method of detecting wastewater treatment plants (WWTP) accurately and automatically via remote sensing is unavailable. In this paper, we provide a solution to this task by proposing a novel joint deep learning (JDL) method that consists of a fine-tuned object detection network and a multi-task residual attention network (RAN). By leveraging OpenStreetMap (OSM) and multimodal remote sensing (RS) data, our JDL method is able to simultaneously tackle two different tasks: land use land cover (LULC) and WWTP classification. Moreover, JDL exploits the complementary effects between these tasks for a performance gain. We train JDL using 4,187 WWTP features and 4,200 LULC samples and validate the performance of the proposed method over a selected area around Stuttgart with 723 WWTP features and 1,200 LULC samples to generate an LULC classification map and a WWTP detection map. Extensive experiments conducted with different comparative methods demonstrate the effectiveness and efficiency of our JDL method in automatic WWTP detection in comparison with single-modality/single-task or traditional survey methods. Moreover, lessons learned pave the way for future works to simultaneously and effectively address multiple large-scale mapping tasks (e.g., both mapping LULC and detecting WWTP) from multimodal RS data via deep learning.

Keywords: GeoAI; OpenStreetMap; SDG 6; multi-task learning; multimodal; object detection; volunteered geographic information; wastewater treatment.

Fig. 1

An overview of the proposed method of joint deep learning. There are mainly three input data: (a) VHR image; (b) OpenStreetMap data; (c) Sentinel-2 MSI.

Fig. 2

Illustrations of the fine-tuned SSD object detection network. Besides the network architecture, (a)-(c) describe the tiling strategy and training objective: (a) SSD requires ground truth boxes and the satellite image as training inputs; (b) 16 $\times$ 16 feature map and default boxes matched with small WWTP; (c) 8 $\times$ 8 feature map and default boxes matched with large WWTP. Based on the matched boxes, the conf and loc losses can be derived.

Fig. 3

Illustration of the Attention Block used for the multi-task classification of WWTP and LULC.

Fig. 4

Overview of the study area and training data. Left: the distribution of OSM WWTP features together with reference features in the test area around Stuttgart; Right: the VHR training and testing WWTP samples and Sentinel-2 MSI of the test area.

Fig. 5

Detailed architecture of the multi-task RAN-56 and its number of parameters.

Fig. 6

Detection Results of WWTP In Four selected Area with Different Methods.

Fig. 7

LULC classification map of four selected areas with different methods.

Fig. 8

Two major results of the JDL method: Left: the LULC classification map produced by the multi-task RAN; Right: the map of clustered WWTP sites with the WWTP bounding boxes detected by the JDL method and their geocoding addresses.