As carbon emissions become increasingly concerning, electric vehicles (EVs) have emerged as a key alternative to fossil fuel-powered transportation. A crucial parameter in EVs is the state of charge (SoC), which represents the remaining battery energy. As SoC cannot be directly measured, deep learning methods, particularly 1D convolutional neural networks (1D-CNNs), are widely used for estimation. However, conventional pooling layers in CNNs, such as max and average pooling, may lead to information loss and negatively impact prediction accuracy. To address this, this study proposes the integration of wavelet transform (WT) into the pooling layers to enhance feature extraction and improve SoC estimation performance. The study utilizes the LG 18650HG2 battery dataset from McMaster University, covering various driving cycles at temperatures between −25 °C and 40 °C. Data preprocessing included the removal of irrelevant segments and adjusting the sampling frequency to improve model training efficiency. SoC estimation performance was evaluated using mean squared error (MSE), and computational efficiency was analyzed for real-time applicability. Experimental results show that the WT-based pooling method outperforms conventional pooling. In the Mix6 cycle at 25 °C, the proposed method achieved an MSE of 0.00365 while maintaining stable performance across different temperatures. Although WT pooling increases computational complexity on GPUs, it performs efficiently on CPUs, making it suitable for real-time applications. Future studies may focus on optimizing computational efficiency and extending this approach to different battery chemistries and driving conditions for improved robustness.