Compressed sensing for object detection using dictionary learning

Abstract

Reliability of data transmission is of critical concern especially in complicated networks like the underwater sensor networks. To mitigate this issue, a data acquisition model is proposed in which alternative data forms viz. image or audio are used and the desired object is detected. Sparse coding is executed by dictionary learning for both types of data. Compressed sensing is applied on the incoming signal based on the type of data that is being accepted. This paper applies the dictionary learning method to denoise a SONAR image to which Gaussian noise is added. The Douglas-Rachford (DR) algorithm is used to perform sparse coding on the learned dictionary of each patch from the numerous patches extracted from the image. For compressed sensing of audio, a set of short time Fourier transforms (STFTs) was computed with multiple windows as an input to the Gabor Transform. Using Basis pursuit, this paper implements dictionary to denoise an audio signal. The results are compared with other methods for performance evaluation. The implementation results and empirical studies suggest that this is a reliable technique for object detection.