MPCM-Net: A Multiscale Network That Integrates Partial Attention Convolution With Mamba for Ground-Based Cloud Image Segmentation

Ground-based cloud image segmentation is a critical research domain for photovoltaic (PV) power forecasting. Current deep learning (DL) approaches primarily focus on encoder-decoder architectural refinements. However, existing methodologies exhibit several limitations: (1) they rely on dilated convolutions for multi-scale context extraction, yet fail to leverage inter-channel interoperability and partial feature efficacy; (2) implementations of attention-based feature enhancement frequently compromise the equilibrium between accuracy and throughput; and (3) the decoder modifications often fail to re-establish global interdependencies among hierarchical local features, thereby constraining inference efficiency. To mitigate these challenges, we propose MPCM-Net, a Multi-scale network that integrates Partial attention Convolutions with Mamba architectures to enhance segmentation accuracy. Specifically, the encoder incorporates a multi-scale partial attention convolution (MPAC), which comprises: (1) a multi-scale partial convolution block (MPC) with partial channel module (ParCM) and partial spatial module (ParSM) that facilitating global spatial interaction across multi-scale cloud formations, and (2) a multiscale partial attention block (MPA) combining partial attention module (ParAM) and ParSM to extract discriminative features with reduced computational complexity. On the decoder side, a multi-scale Mamba block (M2B) is employed to mitigate contextual loss through a spatial-semantic hybrid domain (SSHD) that maintains linear complexity while enabling deep feature aggregation across spatial and scale dimensions. Furthermore, we introduce and release a dataset incorporatingComplex-Scale variations, Radiative properties, and Color attributes (CSRC), which is a clear-label, fine-grained segmentation benchmark designed to overcome the critical limitations of existing public datasets. Extensive empirical analysis on CSRC demonstrates the superior performance of MPCM-Net over state-of-the-art methods, achieving an optimal balance between segmentation accuracy and inference speed. The dataset and source code will be available at https://github.com/she1110/CSRC.