Deep CNN: Classification of Known Classes and Unknown Classes Based on Softmax Prediction Probabilities

Detecting unknown classes in an image classification is one of the key challenges deep learning developers confront. The idea of image recognition is built on deep learning, which can process enormous amounts of unstructured data by initializing weights and biases in neural networks with random valu...

Full description

Saved in:
Bibliographic Details
Published in2023 4th International Conference for Emerging Technology (INCET) pp. 1 - 11
Main Authors Prakash, D. Bhanu, Kumar, K. Arun, Rishitha, M.
Format Conference Proceeding
LanguageEnglish
Published IEEE 26.05.2023
Subjects
Convolutional Neural Network
Convolutional neural networks
Deep learning
Face recognition
Flavia Leaf Dataset
Image recognition
Neural networks
Real-time systems
Softmax
Swedish Leaf Dataset
Training
Unknown class
Online AccessGet full text

Cover

More Information
Summary:Detecting unknown classes in an image classification is one of the key challenges deep learning developers confront. The idea of image recognition is built on deep learning, which can process enormous amounts of unstructured data by initializing weights and biases in neural networks with random values. By using Convolutional Neural Network (CNN) algorithm it is possible to identify between known classes (labeled data) and unknown class with open-set classification. The datasets Flavia Leaf Dataset and Swedish Leaf Dataset are considered for training and datasets that are not a part of training set are also considered for testing. The number of accurate and inaccurate predictions for each class is calculated using the confusion matrix for each class along with performance indicators such as accuracy.
DOI:10.1109/INCET57972.2023.10170405