SEP Activated GoogLeNet Architecture for Electronic Device Classification
Rangaraajan Muralidaran, Ravi Kumar Jatoth · 2025
The research paper presents a tailored GoogLeNet architecture with a novel SEP activation function for image classification tasks in the electronic devices domain. The model incorporates a refined Inception module and undergoes extensive preprocessing for improved feature extraction. The performance evaluation, conducted on a dataset of 1607 electronic device images (resistors, capacitors, transistors, pulse-generators, multi-meters, LEDs, potentiometers, fuses, ICs and breadboards), showcases the model's proficiency in classifying diverse categories of electronic devices. The model architecture, commencing with a 7x7 convolutional layer and progressing through max-pooling, 1x1 convolutions, and custom Inception modules, culminates in an output layer yielding class probabilities. The model achieved an accuracy of 93.04%.