Effective Computer Vision Approach for Surveillance Systems to Detect Stone Crushers

The increase in stone quarrying activity within the Khordha district of Odisha has led to a rise in associated businesses, particularly in stone-crushing operations. The unauthorized operation of these stone-crushing machines results in a significant increase in dust particles in the atmosphere, creating a concerning scenario that poses various health risks for the surrounding villagers. These establishments are spread over a vast geographical area of the district, so it poses challenges for the district authorities to conduct close monitoring. To facilitate aerial surveillance for this geographic location, this study presents the development of a custom model utilizing YOLO (You Only Look Once), a single-stage object detection algorithm grounded in convolutional neural networks, to precisely identify the stone-crushers from complex backgrounds. This model was assessed on a novel dataset comprising 74 high-resolution remotely sensed images obtained by an unmanned aerial vehicle or drone from three distinct geographic regions of the Khordha district. The object detection capabilities of this model proved that it could accurately identify the stone-crushing units with up to 0.99 mAP (Mean Average precision) on our novel high-resolution imagery dataset. The findings of this study indicate that the proposed methodology has the potential to develop aerial surveillance systems while maintaining a good balance between speed and accuracy.