ODinMJ: A red, green, blue-thermal dataset for mountain jungle object detection

Mountain jungle is a key landscape and an important component of the Earth’s surface, with mountains covering approximately 24% and jungles covering about 31% of the global land area. This ecosystem covers nearly half of the world’s biodiversity hotspots. And the relationship between mountain jungle and human society is very close, therefore mountain jungle is of significant research importance. Due to jungle canopy occlusion and low resolution, satellite remote sensing cannot meet the perception needs of mountain jungle. And unmanned aerial vehicles (UAVs) are the most promising jungle observation platform due to their temporal and spatial flexibility, high resolution sensor-carrying capabilities. Object detection is the fundamental technology for various downstream applications. Existing methods, due to the lack of datasets of mountain jungle scenes, cannot adapt well to the complex characteristics of mountain jungle scenes, resulting in the low detection accuracy. Therefore, UAV datasets tailored to mountain jungle scenes are crucial for advancing the development of mountain jungle remote sensing. In this article, we introduce an RGB-T object detection dataset for mountain jungle for the first time, named ODinMJ. The dataset was captured at different time periods and locations, covering various special environmental conditions such as different lighting conditions, slope variations, occlusions, and camouflage factors. We evaluate the performance of various representative object detection models on the ODinMJ dataset to serve as a benchmark for the development of this field. Furthermore, this article discusses insights and future research directions on mountain jungle remote sensing. The dataset is available at
https://github.com/KustTeamWQW/ODinMJ-RGB-T-Dataset
.