Signature Library Creation from Satellite Imagery for Automatic Target Recognition

The development of efficient algorithms for automatic target recognition (ATR) and classification from satellite imageries are of great challenges and are of particular importance for military applications. These algorithms are crucial for extraction of various types of military targets and their geo-location from varying resolution panchromatic, multispectral satellite images. Traditionally the signature libraries of military targets play a key role in determining the accuracy and performance of ATR algorithms. The signature libraries are comprised of target chips (template) or set of features or templates and features both in a hybridized ATR method. In this work, we have developed a unique method for creating exhaustive sets of signature libraries for variety of military targets. This work demonstrates that using our method of creating signature libraries greatly helps in enhancing the ATR results in a hybridized method and to achieve an accuracy level of over 95% in detection result.