Deep learning based glance of real world scenes through decision tree

Abstract

How do computers learn to recognize scene instantly? Humans classify scenes based on local features along with context and also compared with other features. A scene can be described in multiple ways and include details about objects, regions, geometry, location, activities, and even nonvisual attributes. We present a class of efficient streamlined SSD Mobilenet architecture, uses depthwise separate convolution for strong performance on COCO dataset to utilized to recognize real-world objects in Indoor-Outdoor Scene Images. In the scene analysis, the spatial properties are evaluated to generate the final linguistic interpretation and improve classification performance. In this paper, we have proposed an outdoor scene analysis system, by which decision tree rules can be learned to recognize scene categories. The major advantage of this work is by using low-level features, minimal training data, maximum accuracy of prediction analysis. This proposed methodology has been work well to overcome the outdoor scene challenges like lighting conditions and shape complexity. Scene understanding has a great impact on computer vision due to perceiving, analyzing, and interpreting the visual scene which leads to new inventions. In this paper, we present the methodology to recognize the real world scenes in different categories. This approach takes into developing a sequential hierarchy of identifying objects in the scene followed by combining spatial relationships and important scene properties.