Identification and Counting of Sugarcane Seedlings in the Field Using Improved Faster R-CNN

Sugarcane seedling emergence is important for sugar production. Manual counting time-consuming and hardly practicable large-scale field planting. Unmanned aerial vehicles (UAVs) with fast acquisition speed wide coverage are becoming increasingly popular in precision agriculture. We provide a method based on improved Faster RCNN automatically detecting sugarcane seedlings using photography. The Sugarcane-Detector (SGN-D) uses ResNet 50 feature extraction to produce high-resolution expressions provides an attention (SN-block) focus the network learning channels. FPN aggregates multi-level features tackle multi-scale problems, while optimizing anchor boxes size quantity. To evaluate efficacy viability of proposed technology, 238 images were taken from air unmanned vehicle. Outcoming average accuracy 93.67%, our outperforms other commonly used detection models, including original R-CNN, SSD, YOLO. In order eliminate error caused by repeated counting, we further propose de-duplication algorithm. highest reached 96.83%, whilst mean absolute (MAE) 4.6 when intersection union (IoU) was 0.15. addition, software system developed automatic identification cane seedlings. This work can accurate data, thus support farmers making proper cultivation management decision.