<![CDATA[The successful outcome of forest rehabilitation governance projects in Indonesia remains debatable due to a lack of convincing evidence, including the use of manipulable geotagging camera photos. Conversely, advancements in deep learning and machine learning for object detection have shown significant promise, yet their application often remains challenging for local communities and governments due to limited technical expertise. This research investigates the use of high-spatial-resolution UAV imagery integrated with deep learning techniques to identify planting holes and seedlings—key indicators of forest rehabilitation progress. Our study aimed to enhance accountability in forest rehabilitation governance projects by developing a user-friendly and government-aligned model that simplifies application for non-experts. By leveraging eCognition software, we customised learning templates and implemented parameter optimisations to address environmental complexities, such as shadows and dense grass, which often lead to misclassification. Applied to a 200-hectare rehabilitation site in Baringeng Village, Bone Regency, South Sulawesi, our method achieved an accuracy of 85.97% for the planting hole class and 82.15% for the planting seedlings class. This iterative refinement process revealed the potential for broad applicability in forest monitoring projects with tight timelines and the need for independent or third-party assessments. Our findings underscore the value of integrating UAV imagery and automated machine learning workflows for cost-effective and scalable forest rehabilitation monitoring while identifying limitations and opportunities for future research in larger and more diverse environments.]]>
