“‘latex Recent advances in remotely operated vehicle technology and automated processing of visual data through deep-learning approaches have enabled us to track long-term ecological trends at marine rock walls. Here, we trained a deep-learning based object-detection model to classify prominent benthic invertebrate fauna on a slope/wall-section of the Koster fjord, part of the Swedish marine protected area Kosterhavet National Park. The model was applied to footage of the study site from 1997-2023, from which depth ranges and relative abundances of 17 invertebrate taxa were extracted, generating 72,369 occurrence records. The object-detection model was deemed reliable for its purpose with modeled depth distributions aligning with previously documented occurrences. Community structure was found to change along the study site’s depth gradient, with a higher taxon diversity at greater depths. Significant temporal increases in overall abundance across all depths were found in eight taxa and significant decreases in five taxa. The overall community structure shifted toward a higher abundance of small, heat-tolerant suspension-feeders. Temperature preference and size were found to be significant drivers behind taxon-specific abundance change. The documented loss over time of large, heat-sensitive taxa suggests that ongoing temperature increases are a likely cause for the altered community structure. However, a widespread trend of increasing abundance was noted throughout the remaining community, including species sensitive to trawling. This suggests that while species sensitive to climate change may disappear from the area, the remaining benthic community benefits from the protection measures in the national park. Our study demonstrates the application potential of video surveillance and deep learning technology, and we recommend the implementation of standardized video monitoring in adaptive management of marine ecosystems.