The Synthetic Aperture Radar (SAR) is a microwave active imagery system that has been largely used due to its possibility of day-and-night operation in all weather conditions. The SAR system generates images by the coherent processing of the scattering signals; this results in a scene texture that has an undesired multiplicative speckled noise, drastically reduces the ability to distinguish the features of the classes . The rejection of the speckle noise motivated many works where ANN algorithms have been applied to SAR imagery classification . Artificial Neural Network (ANN) algorithms have been increasingly applied to remote sensing for image classification in the last years .
SAR images have found many applications in the field of Automatic Target Recognition (ATR). Target detection is a signal processing problem whereby one attempts to detect a stationary target embedded in background clutter while minimizing the false alarm probability.
The rapid increase of ANN applications in remote sensing imagery classification is mainly due to their ability to perform equally or more accurately than other classification techniques . In a general way, the major advantages of the neural network method over traditional classifiers are:
• Easy adaptation to different types of data and input configuration,
• Simple incorporation of ancillary data sources, as textural information, which can be difficult or impossible with conventional techniques,
• Does not use or need a priori knowledge about parameters of distributions. ANN algorithms find the best nonlinear function, in the optimal case, between the input and the output data without any constraint of linearity or pre-specified nonlinearity which is required.
It has also been shown that supervised neural network classifiers (NC) have outperformed unsupervised methods because the last one utilizes no a priori class information .
Overview of current literature
Various techniques for the classification of SAR images exist, and have been compared using various platforms/benchmarks. Different detection schemes have been proposed and investigated for target detection in SAR images. However, they always utilized the generalized likelihood ratio test (GLRT) as the detection statistics, which is a special case of the quadratic discriminant function in the feature space.
The design of a target detector can be divided into two phases: signal representation and detector formulation. Through signal representation, targets are described more prominently than clutter in terms of the extracted features, and the detector can be formulated to detect these features. The two phases are so intertwined, that it’s unavoidable to go back and forth between these two phases before any powerful detector is derived. Current target detection approaches in this research field utilize the characteristic reflections of metallic objects. There have been many...