Benchmark of State-of-the-art Methods for Plantation Object Identification and Counting

With rising global demand for palm oil, the production of the palm oil needs to be taken seriously. Field maintenance plays an important role in affecting palm oil production. However, since the covid-19 pandemic, severe labor problems have increased the difficulty of oil field maintenance. In this study, the concept of precision agriculture is applied to proposed an effective classification method for palm oil tree management. A few object detection models are built and evaluated to ease the field maintenance processes. The four state-of-the-art models chosen are Faster RCNN Resnet50 with FPN, Faster RCNN Resnet101 with FPN, RetinaNet Resnet50 and YOLOv7. YOLOv7 score the highest mAP of 93.10%, FRCNN-R50-FPN and FRCNN-R101- FPN models have best overall Fl-score. This can be summarized by the fact that the latest object detection available is mature enough to be used widely, not only in the palm oil field.