Across New Zealand and Australia sea urchin overgrazing has become a focal issue for kelp forest ecosystems. Greater resources are being put towards understanding impacts of native and range-expanding species, while kelp restoration efforts are also increasing. Evaluating overgrazing impacts and restoration effectiveness requires robust monitoring data. Benthic imagery is ideal for this monitoring as it can be acquired rapidly over ecologically relevant spatial scales. However, the resultant datasets are typically large and laborious to manually annotate, slowing data analysis and inference making. Here we outline the development of a machine learning (ML) toolkit designed to automate benthic image annotation associated with kelp forest ecosystems. The open-source toolkit contains an urchin detector and habitat classifier which work in tandem to identify key urchin species and habitat forming biota. This information helps classify images into several habitat types allowing rapid inference over surveyed areas and assessment of any temporal changes. Annotation, ML integration and validation have been conducted through Squidle+, with data sourced from imagery platforms across New Zealand and Australia making it robust and widely applicable. Incorporation within existing and future projects will significantly improve data analysis timeframes, enabling greater understanding of the extent of overgrazing impacts and restoration success.