A MASK-RCNN Based Approach Using Scale Invariant Feature Transform Key points for Object Detection from Uniform Background Scene

Object identification using deep learning in known environment gives a new dimension to the research area of computer vision based automation system. As it uses supervised learning technique using Convolution Neural Network (RCNN) it helps automation software tools and machines to detect and identify objects using vision based systems. One of RCNN technique known as Mask-RCNN has been applied in this proposed design and this paper presents a novel approach to object detection problem using Big Data storage for large set of features based data. Earlier work Faster Region-based CNN has led to the development of a state-of-the-art object detector termed as Mask R-CNN. Some samples of solid material objects used in refractory industry have been taken as input images. In our experiment the SIFT based features have been implemented and trained using filter and convolution operation. In addition to improved accuracy, pixel-level annotation (annotating bounding boxes is approximately an order of magnitude which is quicker to perform). The model is retrained to perform the detection of four types of metal objects, with the entire process of annotation and training for the new model per solid block. A key benefit of feature based Mask-RCNN approach is high precision (~94%) in classification and minimized feature points with SIFT key points.