Object detection and classification: a joint selection and fusion strategy of deep convolutional neural network and SIFT point features

Abstract

In the area of machine learning and pattern recognition, object classification is getting an attraction due to its range of applications such as visual surveillance. In recent times, numerous deep learning-based methods are presented for object classification but still, set of problems/concerns exists which reduce the overall classification accuracy. Complex background, congest situtaions, and similarity among different objects are few challenging issues. To tackle such problems, we propose a technique by using deep convolutional neural network (DCNN) and scale invariant features transform (SIFT). First, an improved saliency method is implemented, and the point features are extracted. Then, DCNN features are extracted from two deep CNN models like VGG and AlexNet. Thereafter, Reyni entropy-controlled method is implemented on DCNN pooling and the SIFT point matrix to select the robust features. Finally, the selected robust features are fused in a matrix by a serial approach, which is later fed to ensemble classifier for recognition. The proposed method is evaluated on three publically available datasets including Caltech101, Barkley 3D, and Pascal 3D and obtained classification accuracy of 93.8%, 99%, and 88.6% - clearly showing the exceptional performance compared to existing methods.