Volume 19, No. 4, 2022
Hybrid Methodology With Convolutional Neural Networks And Data Augmentation For Aerial Image Object Detection For Illegal Minery
JIMMY ANDERSON FLÓREZ ZULUAGA , JOSÉ DAVID ORTEGA PABÓN , JUAN CAMILO DÍAZ PAZMIÑO , JHON FREDY ESCOBAR SOTO
Abstract
Automated classification of aerial remote sensing images is essential in automation processes and machine learning fields. This article describes an experimental technology to detect objects of interest in multispectral aerial images. This technology aims to identify areas with illegal mining, an issue that affects both the natural environment and security in Colombia; this is accomplished through the combination of digital image processing (DIP) techniques. DIP is used to filter objects of non-interest, based on their shape and color, from convolutional neural network systems based on a pre-trained MobileNetV2 model; dataaugmented metrics and data without augmented metrics were then used for the detection of the illegal mining phenomenon. With classification systems based on artificial intelligence, DIP could improve the efficiency of the autonomous processing of high-resolution input images in terms of time and identification of false positives. The application of digital image processing PDI with classification systems based on artificial intelligence could improve the efficiency in the results of autonomous processing of high-resolution input images, that could be used for detection process automatization.
Pages: 532-554
Keywords: Aerial images image processing, satelital multiespectral image processign, image classification, mobilnetv2, Image digital processing, illegal mining, artificial intelligence.