Citrus Fruit Detection


In-field fruit detection and counting are crucial tasks for agricultural automation. For example, they can be used to reduce routine farming and breeding activities as well as provide insightful estimates for harvest and forthcoming growing seasons. Moreover, accurate fruit detection enables the possibility of robotic harvesting, which has the potential to eliminate one of the most labor-intensive processes for growers. Many imaging and sensing technologies have been used for detecting fruit such as hyperspectral, laser scanning, thermal, and RGBD sensors, yet the most common technology is the standard RGB camera. Although conventional RGB cameras are widely accessible, they present several challenges for in-orchard fruit detection such as variation in appearance, irregular lighting, and severe occlusion. Recent works have used deep learning approaches to overcome these challenges. Nonetheless, due to the lack of standardized benchmark datasets for agricultural automation, it is difficult to compare existing methods with each other.


  • We created a high-quality citrus fruit image dataset along with baseline performance benchmarks on multiple state-of-the-art object detection algorithms


  1. J.A. James, H.K. Manching, M.R. Mattia, K.D. Bowman, A.M. Hulse-Kemp, and W.J. Beksi, "CitDet: A Benchmark Dataset for Citrus Fruit Detection," 2024.
    Preprint  •   Website  •   Citation