Image Classification using Transfer Learning and Convolutional Neural Networks (CNN)
The amount of data in the form of images that have been collected, both those that have been disseminated as image data that can be processed or used as a guide for other researchers both for developing research and for designing new research, still requires a deep learning approach to assist in conditioning the image data. This is intended because the image data has a variety of conditions, it can only be a collection from the internet, or it could also be the data from the photos themselves or their captures, which have different scales and resolutions. Therefore, based on the studies that have been carried out, there are still some deficiencies, such as dataset problems, both in the number of datasets to the variations of the dataset used, feature extraction used, and preprocessing carried out to the architectural model building. Convolutional Neural Network (CNN) is a deep learning algorithm representing the development of multi-layer perceptrons (MLP). CNN can classify labeled data that describes its neurons in a two-dimensional format. Therefore, CNN is one of the most popular and accurate models implemented in image classification.
Convolutional Neural Networks (CNN) can detect and recognize objects in an image. It can also outperform traditional computer vision and pattern recognition methods, such as object detection, classification, image segmentation, and text recognition. In addition, Transfer learning is a deep learning technique that uses a model trained on one problem to solve another problem. Transfer learning enables deep learning training with a small number of samples with high accuracy. Therefore, a classification model built using CNN by combining two Transfer Learning models, namely VGG-16 and Xception, can make the dataset more varied by using data augmentation techniques to enrich the dataset that is owned and is expected to avoid overfitting cases—combining the CNN model with Transfer Learning to increase the accuracy results previously obtained from CNN model training.
Read also : How to Choose The Best Machine Learning Algorithms and Tools
Image processing is the study of algorithms that take images as input and return images as output. Today, image processing is a primary tool in many scientific fields, including computer science, electrical engineering, robotics, physics, chemistry, environmental science, biology, and psychology. Image processing is a subcategory of signal processing.
The input of signal processing is an image, and the output is also an image or a function associated with it. The image processing technology is developing rapidly, and image processing has been widely studied in information technology and engineering fields.
Read also : Voice Recognition Analyisis Using MFCC
Digital image processing (Image Manipulation) is manipulating the pixels of a digital image for a specific purpose. An image is considered digital if the resulting image is output from a computer, camera, scanner, or another electronic device. Digital image processing is carried out using a computer algorithm where digital images are presented as matrices, digital image processing is primarily the manipulation of matrix elements into pixels—the difference between the traditional machine learning approach and the transfer learning approach. The image shows two different data sets, cars, and buses. The dataset has two distinct tasks, namely classifying passenger cars and buses. Traditional machine learning requires practice over time to learn a task, whereas transfer learning requires training a pre-trained model on a new target domain with a smaller training sample.
