Detecting motion-induced blur palm tree for oil palm automation in online system settings

Efficient palm tree detection forms the bedrock of automation systems within oil palm plantations, serving as the initial step before subsequent processes unfold. This encompasses a spectrum of applications—from targeted spraying in automated systems, which curb excessive chemical runoff, to pivotal tasks like harvesting, weed management, and yield estimation. The challenge, however, emerges when visual sensors capture images while in motion. This paper delves into the imperative of identifying palm trees obscured by the blurring effects of camera movement atop mobility platforms within online system settings. The YOLO-based detector is used as the backbone of this research. The focus rests on training these detector models to identify palm trees not only in sharp form but also in distorted states, such as motioninduced blur. Training involves three distinct models employing different training datasets: one contains solely sharp palm tree images, another features a blend of sharp and blurred objects and a third that segregates the two categories. The models' performance is evaluated by detection accuracy and inference time. Empirical findings validate the efficacy of our approach in accurately pinpointing motion-blurred palm trees, underscored by its potential for seamless real-time integration. This study greatly helps improve the accuracy and reliability of systems that detect palm trees. It's crucial for the important job of analyzing visual data online.