Over the recent past, CNN-based change detection methods have achieved significant advancement. However, there still exist some limitations: (1) Large-kernel convolutions introduce background noise, reduce edge sharpness, and increase parameter redundancy, complicating mobile deployment; (2) Dilated convolutions expand receptive fields but lose local texture details due to feature sparsity; (3) Most methods inadequately model global-local dynamic interactions, leading to high false detection rates. To address these challenges, we propose a Multi-Scale Fusion Lightweight Change Detection Network Based on Rich Information Extraction (MSFLNet). The Lightweight Global Dynamic Feature Extraction Module (LGDM) enhances the capture of multi-directional features by decomposing large kernel convolutions and achieves cross-scale dynamic feature fusion in tandem with the Global Enhanced Multi-Scale Attention (GEMA). The Lightweight Spatio-Temporal Local Feature Extraction Module (LSTLM) mitigates background interference and bolsters the representation of local details via grouped dilated convolutions and feature masked mechanisms. Additionally, a parameter-free attention mechanism is incorporated to model the spectral-spatial joint relationship, based on an energy function, which dynamically amplifies the response of anomalous regions within multi-band images. The Adaptive Feature Fusion Module (AFFM) dynamically adjusts global and local features through efficient channel attention, achieving adaptive fusion of multi-level information and augmenting discriminative power. Experimental outcomes on the LEVIR-CD and CDD datasets, along with ablation studies, demonstrate that the proposed MSFLNet surpasses existing methodologies.