LogoKAN: Leveraging Kolmogorov-Arnold Networks (KANs) and CNNs for Logo Classification with Applications in IC Assurance

Object classification, encompassing logo classification, constitutes a significant challenge within the realm of computer vision, with the recognition of Integrated Circuit (IC) logos introducing distinct complexities attributable to the scarcity of labeled data and the presence of inter-class similarities. This research critically assesses the efficacy of Kolmogorov Arnold Networks (KANs) and Multi-Layer Perceptrons (MLPs) as foundational architectures for Convolutional Neural Networks (CNNs) such as ResNet50 and VGG16, utilizing datasets including CIFAR10, MNIST, FlickrLogos-32, and a novel IC Logo Dataset comprising 284 distinct categories and a total of 2832 images. The findings reveal that KAN-enhanced CNNs consistently achieve superior metrics in terms of accuracy, precision, recall, and F1-score across the majority of datasets, with KAN-ResNet50 attaining the highest performance levels on CIFAR10 (94.11%), MNIST (99.14%), and IC logos (73.52%). In the context of FlickrLogos-32, KAN-ResNet50 and MLP-ResNet50 exhibited comparable performance, with the MLP variant providing marginally enhanced inference times. Likewise, CNNs supported by MLPs manifested competitive performance metrics on CIFAR10 and MNIST, particularly when prioritizing computational efficiency and inference speed. These outcomes elucidate the inherent tradeoffs between accuracy and efficiency, wherein KANs furnish robust capabilities for feature extraction and representation that are particularly advantageous for complex tasks, whereas MLPs facilitate expedited processing for less resource-intensive scenarios. This study accentuates the transformative potential of KANs within CNN architectures for applications necessitating high accuracy and robust generalization, especially in the domain of IC logo classification, while concurrently emphasizing the critical importance of model selection tailored to specific task requirements.