A Review of image Classification and Object Detection on Machine learning and Deep Learning Techniques

• Published in: 2021 5th International Conference on Electronics, Communication and Aerospace Technology (ICECA) pp. 1 - 8
• Main Authors: Sandhya Devi, Ms. R.S., Vijay Kumar, V.R., Sivakumar, P.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 02-12-2021
• Subjects: Classification algorithms; Computer vision; Convolutional Neural Network (CNN); Convolutional neural networks; Deep learning; Image classification; Machine learning algorithms; Network architecture; Object detection; TensorFlow
• DOI: 10.1109/ICECA52323.2021.9676141

One of the known computer vision tasks is image recognition and classification. Computer vision - a field that automates tasks and works with multi - dimensional data, when combined with the power of artificial intelligence mimics human visual system. Image processing and analysis is an integral part of computer vision. Image classification, a vital part in image analysis is used to categorize pixels or classes of interest by extracting information classes from a multiband raster image. Automatic sorting of pixels and analogous multispectral reflectance values are widely used in land cover mapping cases where remote sensing and Geographical Information System (GIS) is integrated. Designing of image classification and processing procedure by considering various factors involved, decides the success of the classification approach being used. This paper provides a summarization of various classical machine learning based classification approaches and the major advanced deep learning based classification techniques, issues affecting the success of each classification approach and its applications. Convolutional Neural Network (CNN) is the most popular deep network architecture used to analyze visual imagery. This paper discusses waste object detection, a part of the waste segregation process based on CNN hierarchical image classification approach. Here, the system is able to detect the object and provides the relative match percentage of the object being detected. Open source software libraries such as Tensorflow and Spyder is used for this process