Sentiment analysis has gained considerable attention from scholars, resulting in significant achievements and applications. However, traditional single-modal sentiment analysis focuses only on text or image data, while users often share text and images on social media. This study proposes a multi-modal sentiment analysis model that integrates an attention mechanism to leverage the complementary and comprehensive information from image and text features. Textual features are extracted using a pre-trained BERT-base model and an LSTM to enhance semantic understanding and generate coherent, context-aware representations with attention mechanisms. Meanwhile, visual features are obtained through CNNs and attention mechanisms. These multi-modal features are then combined using a fusion layer with a linear transformation and a ReLU activation function. Experimental results on the Single and Multiple datasets demonstrate that the proposed approach outperforms existing methods.