In this paper, we suggest an unmanned aerial vehicle (UAV) system on industrial chimneys using UAV’s high mobility. The UAV follows along the waypoints given by operators and recognizes the front obstacles by using the stereo camera and re-plans the path. By applying a chimney temperature state (fire detection) classifier using Support Vector Machine (SVM), we solved the issue of a lack of data to detect. The SVM classifier is trained by small-scale experimental datasets. To apply different scale data for the SVM classifier, we extracted vector with Histogram of Oriented Gradients(HOG) and checked the appropriate scale by estimating similarity (Wasserstein distance) of varying scale reference images. The autonomous flying tests were conducted at the industrial environment and the feasibility of the chimney inspection algorithm has been done with lab-scale experiments.