Given the limitations of visual sensors in beach environments, an innovative design for the sound source recognition module of beach search and rescue robots is proposed. The audio recorded by acoustic sensors in beach environments contains a large amount of noise such as the sound of waves, which greatly interferes with the recognition of the sound source of the person in distress. To address this, the complex audio in the beach environment is analyzed, and a wave sound signal filter is designed based on the acoustic signal characteristics of each component to effectively remove the invalid components from the original audio. At the same time, to address the issue of the small number of beach acoustic datasets and the difficulty in recording audio files under various sea conditions, a high-precision beach distress call dataset is created based on real-world beach environment recordings and human voice audio datasets. After filtering the audio dataset through the wave sound signal filter, the Mel spectrogram is obtained and passed into a lightweight convolutional neural network for training. Simulation results show that the recognition success rate of the model after signal filtering for the person in distress in the beach environment has increased from 90.67% to 97.42% compared to the model without signal filtering. Notably, compared with traditional approaches, this method achieves a recognition success rate comparable to ResNet18 while demonstrating a substantial improvement in recognition speed.