LSCS-Net: A lightweight skin cancer segmentation network with densely connected multi-rate atrous convolution
Abstract: Highlights•A novel skin lesion segmentation network with only 401K parameters is proposed.•A novel feature block proposed to enhance the robustness and efficiency by combining multiple scales of Information.•Proposed a dense multi-rate atrous convolution block to enhance global semantic features.•The LSCS-Net avoids filter overlap by using optimal filters and depth-wise separable convolutions•Significantly reducing training time and parameters.•The LSCS-Net shows competitive performance and resource efficiency, making it suitable for skin cancer segmentation with limited resources.
Loading