G-Net Light: A Lightweight Modified Google Net for Retinal Vessel Segmentation

In recent years, convolutional neural network architectures have become increasingly complex to achieve improved performance on well-known benchmark datasets. this research, we introduced G-Net light, a lightweight modified GoogleNet with filter count per layer reduce feature overlaps, hence reducing the complexity. Additionally, by limiting amount of pooling layers in proposed architecture, exploited skip connections minimize spatial information loss. The suggested architecture is analysed using three publicly available datasets for retinal vessel segmentation, namely DRIVE, CHASE and STARE light achieves an average accuracy 0.9686, 0.9726, 0.9730 F1-score 0.8202, 0.8048, 0.8178 CHASE, datasets, respectively. state-of-the-art outperforms other segmentation fewer trainable number parameters.