LightNet: A Lightweight Neural Network for Image Classification
Akshay Kumar Sharma, Byungho Kang, Kyung Ki Kim · 2021
Image Classification is widely used in the field of computer vision that focuses on classifying the object in a given image. Lately, image classification techniques are not only restricted to computer applications but are also famous for edge devices. Convolutional Neural Networks play a crucial role in building a good image classifier. However, in order to achieve high accuracy, CNN algorithms lead to use a large number of layers that results in increasing the number of parameters and makes it difficult to implement on edge devices. To overcome this problem, a lightweight image classifier "LightNet" is proposed in this paper that makes use of different scales of receptive fields to extract more feature maps with fewer parameters. To examine the efficacy of the proposed lightweight classifier, it is tested on the CIFAR-10 dataset and got 90% accuracy by using only .26M parameters, which shows that the proposed Lightnet is very effective to be implemented on edge devices.