A figure of 33,000 search and rescue (SAR) incidents were responded to by the UK’s HM Coastguard in 2020, and over 1322 rescue missions were conducted by SAR helicopters during that year. Combined with Unmanned Aerial Vehicles (UAVs), artificial intelligence, and computer vision, SAR operations can be revolutionized through enabling rescuers to expand ground coverage with improved detection accuracy whilst reducing costs and personal injury risks. However, detecting small objects is one of the significant challenges associated with using computer vision on UAVs. Several approaches have been proposed for improving small object detection, including data augmentation techniques like replication and variation of image sizes, but their suitability for SAR application characteristics remains questionable. To address these issues, this paper evaluates four float detection algorithms against the baseline and augmented datasets to improve float detection for maritime SAR. Results demonstrated that YOLOv8 and YOLOv5 outperformed the others in which F1 scores ranged from 82.9 to 95.3

Remote sensing technologies have been increasingly crucial to support policy-makers in achieving their ecological strategies. The data provided by such technology can estimate the bioenergy source production rate and monitor deforestation. This work participates in the cause by contributing an aerial dataset and developing an intelligent tree-detection system usable for counting trees with the bioenergy potential. Low-altitude flying units have been vastly used for such a purpose due to their ability to capture high-quality data from distant locations. Despite these potentials, collected images that compose a dataset are often characterized by imbalanced distribution among classes. The class disproportion can affect the overall model performance, as it severely deprives key features of under-represented classes. This study proposes data-level approaches that adopt and extend prior sampling algorithms for object detection problems. The devised techniques try to reduce the number of redundant outputs obtained from sampling methods and reduce the iteration required to achieve the target imbalance ratio by employing a systematic flow. In such a process, the class distribution of an original dataset is used as a guideline for selecting candidates for subsequent processes. Our results show that the modified dataset can reduce the length of a training process shown by fewer iterations required to achieve the final metrics of its original dataset version and lower training losses in each iteration. Additionally, the modified dataset can improve the F-score ( F 1 ) and precision metric of object detection algorithm by up to 6%.

The Wallacea region of Sulawesi, Indonesia is renowned for its biodiversity and exceptional endemism. Over the last decade, the region is vulnerable to deforestation, degradation and illegal activities. Frequent monitoring in terms of tree counting provides useful information for various stakeholders such as forest management, government institutions, and environmental agencies. Existing monitoring methods include labour intensive manual observations and satellite imaging remote sensing technology. Satellite-based imagery is low resolution, infrequent, and sometimes include cloud cover. To overcome these drawbacks, this research utilises UAV-based high-resolution RGB images processed by machine learning algorithm to detect tree species, i.e., Sugarpalm, Clove, and Coconut. We compared many deep learning algorithms and found that YOLOv5 model is lightweight, easy to use, fast and accurate for tree species identification.