Cepstral Identification Techniques of Buried Landmines from Degraded Images Using ANNs and SVMs based on a Spiral Scan

In this paper new identification techniques for buried landmine objects are presented.  Most of the existing supervised identification methods are based on traditional statistics, which can provide ideal results when sample size is tending to infinity. However, only finite samples can be acquired in practice. In this paper, two proposed learning methods; Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs), are applied on landmine images. The complete identification technique consists of two stages to perform both the training of the input image models and the evaluation of the testing image sets. In the 1st stage, the 2-D images are transformed into 1-D signals by a spiral scan, and then the Mel Frequency Cepstral Coefficients (MFCCs) and polynomial coefficients are extracted from these 1-D signals and/or their transforms. In the 2nd stage, the ANN and SVM are used to match the extracted features in the testing phase to those of the training phase. Experimental results have shown that the proposed techniques are effective with landmines. The best performance has been achieved with features extracted from the Discrete Cosine Transform (DCT) signals using ANN and from the DCT of images contaminated by AWGN and speckle noise and from the Discrete Sine Transform (DST) of images contaminated by impulsive noise using SVM. Finally, we can say that the proposed techniques achieve better performance compared to other techniques.