A Rapid Object Recognition Scenario Based on Improved YOLO-V3 Network

The object recognition is an essential field in computer vision which comprises locating and recognizing objects within a given input image. The basic principle of these techniques is utilized to recognize certain object and detect their location through generating a list of squares to surround the object. This paper proposed a You Only Look Once Version3 (YOLOV3)-MobileNet for recognizing the object. The Caltech101 dataset is utilized in this research and preprocessed through min-max normalization and contrast enhancement techniques. The preprocessed images are fed into the integration of YOLOV3 and MobileNet classification. The YOLOV3 is known for real-time object detection capabilities and using MobileNet makes it more effective. The performance of YOLOV3-MobileNet is estimated through the performance measures of accuracy, precision, sensitivity, specificity and f1-score. The YOLOV3-MobileNet attained high accuracy of 96.82%, precision of 93.66%, sensitivity of 93.47%, specificity of 93.21% and f1-score of 93.08% which is comparatively higher than existing methods like Fuzzy C-means and Saliency-Hierarchical Max-poling (SAL-HMAX).