Detection and Classification of Cetacean Echolocation Clicks using Image-based Object Detection Methods applied to Advanced Wavelet-based Transformations

A new AI research paper presents CLICK-SPOT, a system that automates the detection of whale and dolphin communication signals. Developed by Christopher Hauer, the approach tackles a major bottleneck in marine bioacoustics: the manual labeling of animal sounds like clicks and whistles for behavioral research is prohibitively slow. The system uses deep neural networks (DNNs) to process underwater audio, but with a key innovation. Instead of relying on standard spectrograms created via Short-Time Fourier Transform (STFT), which suffer from a time-frequency resolution trade-off due to the uncertainty principle, CLICK-SPOT uses wavelet-based transformations. Wavelets provide superior time resolution for high frequencies and better frequency resolution for low frequencies, making them ideal for capturing the rapid, complex nature of cetacean echolocation clicks in noisy ocean environments. The transformed audio data is treated as an image, allowing the application of proven computer vision object detection methods to 'see' and classify the clicks. The model was validated on real-world Norwegian killer whale recordings provided by cetacean biologist Dr. Vester, demonstrating its practical efficacy. This work builds on prior systems like ANIMAL-SPOT but refines the feature extraction process. The technical contribution lies in effectively bridging advanced signal processing (wavelets) with modern AI (image-based DNNs) to solve a specific, data-intensive ecological challenge. For researchers, this means moving from painstaking manual review to automated, scalable analysis of acoustic data.