PERFORMANCE EVALUATION OF DUAL CHANNEL OPTOELECTRONIC SURVEILLANCE SYSTEM WITH NEURAL NETWORK INFORMATION FUSION

Abstract

Optoelectronic surveillance systems are widely used in many areas of human activity: in agriculture, medicine, military systems, in search and rescue operations. These systems are designed for space, airborne, ground and maritime applications. They have to work under all climate conditions with absolute certainty under all light and weather conditions. And on the other hand, they have to be cost effective. That’s why optoelectronic surveillance systems use channels which provide complimentary information about the object and background. And for the most applications channels that operate in visible and thermal ranges of optical spectrum are indispensable element of such systems. Optoelectronic surveillance systems have to continuously meet and even exceed today’s performance and reliability requirements. That’s why there is a need to use not only cutting-age technology, but also adaptive signal processing. Information fusion is state-of-the-art technic to improve overall system performance. Numerous image fusion methods have been proposed during several decades but the most promising are neural networks.

Television system, which work in visible range of optical spectrum, and thermal system, which work in long-wave infrared range where chosen for the modeling. Probability of target detection, recognition and identification was used for performance evaluation. Probability of target detection, recognition and identification for separate long-wave infrared and television channels were modeled. Also, probabilities were estimated for the fused data. Information fusion was done with the help of convolutional neural networks. Simulation results showed that probability of target detection, recognition and identification are almost for 6% higher for fused data compared to separate channels.