Assessing the suitability of convolutional auto encoder as an unsupervised tool for image classification in artisanal small-scale gold mining environment
Author(s): Sylvester Akpah, Yao Yevenyo Ziggah and Daniel Mireku-Gyimah
Abstract: Efforts to control or stop illegal Artisanal Small-Scale Gold Mining (ASGM) in Ghana, which is causing significant environmental degradation, have faced numerous challenges because these illegal activities are carried out in remote areas, inaccessible by the current practice of using 4WD vehicles or trekking by foot. This paper sought to assess the suitability of using an Unmanned Aerial Vehicle (UAV) to capture the locations and features of all ASGM sites and use a Convolutional Autoencoder (CAE) to classify the defined sites into legal and illegal ASGM sites. The classification process used by the CAE involved three main stages, namely encoding, latent space learning, and decoding. The encoder accepts the UAV captured images as input, processes the input images to extract salient features and the decoder decodes the salient features to reconstruct the input image and define a site as an ASGM site. To classify a defined ASGM site as legal or illegal, a python program was integrated into the CAE which makes use of known point coordinates of all legal ASGM sites. A site is flagged as illegal if its point coordinates do not match those in the legal ASGM sites database, otherwise, it is a legal site. The performance of the CAE was measured using the following performance metrics: accuracy, precision, recall, and FI-score. The results of the CAE proved superior giving a classification accuracy of 97.52% when compared with the results obtained from other classification algorithms, namely Random Forest (RF) and Support Vector Machine (SVM) with 93.23% and 95.66% respectively. In this paper, it has been demonstrated that UAVs can be used to capture the locations and features of all ASGM sites, which otherwise would have been inaccessible by the use of 4WD vehicles or trekking, and classify the captured location into legal and illegal ASGM sites using a CAE, to facilitate the control and prevention of illegal ASGM in Ghana.
Sylvester Akpah, Yao Yevenyo Ziggah, Daniel Mireku-Gyimah. Assessing the suitability of convolutional auto encoder as an unsupervised tool for image classification in artisanal small-scale gold mining environment. Int J Comput Artif Intell 2022;3(1):26-40. DOI: 10.33545/27076571.2022.v3.i1a.43