Mixed‐decomposed convolutional network: A lightweight yet efficient convolutional neural network for ocular disease recognition

Eye health has become a global concern and attracted broad attention. Over the years, researchers have proposed many state-of-the-art convolutional neural networks (CNNs) to assist ophthalmologists in diagnosing ocular diseases efficiently precisely. However, most existing methods were dedicated constructing sophisticated CNNs, inevitably ignoring trade-off between performance model complexity. To alleviate this paradox, paper proposes lightweight yet efficient network architecture, mixed-decomposed (MDNet), recognise diseases. In MDNet, we introduce novel depthwise convolution method, which takes advantage of dilated operations capture low-resolution high-resolution patterns by using fewer computations parameters. We conduct extensive experiments on clinical anterior segment optical coherence tomography (AS-OCT), LAG, University California San Diego, CIFAR-100 datasets. The results show our MDNet achieves better complexity than CNNs including MobileNets MixNets. Specifically, outperforms 2.5% accuracy 22% parameters 30% AS-OCT dataset.