International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395 -0056
Volume: 04 Issue: 05 | May -2017
p-ISSN: 2395-0072
www.irjet.net
A DEEP LEARNING MODEL FOR IMAGE CLASSIFICATION Aparna R. Rout1, Prof.(Dr.) Sahebrao B. Bagal2 1PG
Student, Department. Of Electronics and Telecommunication, Late G.N Sapkal College of Engineering, Nashik, Maharashtra, India 2Professor, Late G.N Sapkal College of Engineering, Nashik, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract In Image classification we classify image
whose classes are known. In Image Classification, we classify an image into one of the predefined classes or multiple classes at the same time. With the rapid increase of digital photography, image understanding becomes increasingly important. Image semantic understanding is typically formulated as a multi-class or multi-label learning problem [1]. . In traditional supervised learning, an object is represented by an instance (or feature vector) and associated with a class label. Formally, let X denote the instance space (or feature space) and Y the set of class labels. Then the task is to learn a function f : X → Y from a given data set {(x1, y1), (x2, y2), · · · , (xm, ym)}, where xi € X is an instance and yi € Y the known label of xi. Although the above formalization is prevailing and successful, there are many real-world problems which do not fit this framework well, where a real-world object may be associated with a number of instances and a number of labels simultaneously. For example, an image usually contains multiple patches each can be represented by an instance, while in image classification such an image can belong to several classes simultaneously, e.g. an image can belong to mountains as well as Africa [2]. Although this is always not a difficult task for humans, it has proved to be an extremely difficult problem for machines. Image classification is a widely studied problem in the field of Machine Learning for which there are many techniques and algorithms proposed. Deep Learning is one such technique. This work focuses on the application of deep learning algorithms for multi-label, multi-class Image Classification.
into one of the predefined classes. In conventional way, people use different computer vision techniques to extract features from images different machine learning algorithms uses these extracted features to classify the images. Various Supervised machine learning algorithms have been applied to multilabel image classification problems which have also brought successful results. It has become very difficult task to classify the images into interpretative classes. Apart from various learning algorithms the accuracy and performance of the model mostly depends on the trained dataset and the algorithm used. In this paper we have proposed a system to classify the scenery images into different groups of sunset, desert, mountains, trees and sea. In this paper the Current approaches for image classification make essential use of machine learning methods. We focus on deep learning techniques for feature extraction and classification of images. For multi class image classification we created dataset having landscapes scenes of sunset, desert, mountains, trees and sea. For multi label we use natural scenes dataset. In multi label classification problem an instance can have presence of more than one class among the given classes. There are methods to solve multi label classification problem but most of them are based on creating number of binary model equal to the number of labels and this technique is nothing but the Binary Relevance method. In this project, we propose a model which does not require creating multiple binary models instead it has single model which predicts the probabilities of different labels and uses probabilistic threshold values for respective label to convert those probabilities into presence and absence of class/label. Keyword:
Supervised,
Multilabel,
Deep
To summarize the proposed deep learning method using ConvNet for multi label image classification has the following key features compared to the existing methods: 1. A multi stage deep learning framework is designed to do the local discrimination and build local classifier. 2. Convolutional Neural Network is used for feature extraction and the Neural network is used for the classification purpose. 3. CNN is learned in a multi-instance learning fashion using the transfer learning
learning,
machine learning, Binary Relevance Method
1. INTRODUCTION The term image classification refers to the labelling of images into one of a number of predefined categories. Classification is a task to identify the class/category of new instance based on training set
© 2017, IRJET
|
Impact Factor value: 5.181
|
ISO 9001:2008 Certified Journal
|
Page 814