A FUSION CONVOLUTIONAL NETWORK FOR INTELLIGENT ANALYTICS OF AIRBORNE IMAGING

Abstract

Precise airborne imaging and its analysis for outdoor objects suffers from varied object resolutions, cluttered backgrounds, unclear or contaminated surfaces, and illumination conditions. Lately, there have been significant improvements in airborne imaging owing to the use of new and improved sensors and ameliorated unmanned aerial vehicle UAV technologies. However, such improvements call for novel and innovative approaches for image data analysis with matching sophistication. Impending research issues associated with the airborne imaging are data acquisition and augmentation, data communication, accuracy saturation in the image processing algorithms and requirement of information fusion for the precise analysis. This thesis attempts to explore some of these prevalent research questions in the pretext of airborne imaging of outdoor high voltage insulators used in transmission lines. Accurate information about the insulator surface condition is important and is of a high priority since insulator to prevent insulator and subsequent electric failures. A novel Fusion Convolutional Network (FCN) is proposed in this research for potential real-time monitoring of insulators using unmanned aerial vehicles (UAV) as an edge device. A multi-modal information fusion (MMIF) system is developed during this research to analyze and classify possible contaminations present on the electrical insulators. It is shown with evidence that the proposed FCN improves the classification accuracy besides being efficient with respect to the computational time. The proposed FCN is further compared and benchmarked vis-à-vis, conventional classification algorithms such as, IBK, Naïve Bayes, SMO and J48. Potential hardware implementation of the proposed FCN, on emerging edge devices, is also conceptualized and proof of the concept experiments are conducted in silico. Pertinent outcomes of this research can be further extended to other potential applications of airborne imaging.