Sugarcane Internode Dataset: SIDCNet for Detection and Counting of Sugarcane Internodes for Precision Agriculture

This research proposes a custom sugarcane internode dataset. Sugarcane internode count analysis is a crucial parameter that helps experts analyse the cane health status, quality, and quantity of the juice to measure yield and productivity. The conventional sugarcane internode counting methods are done manually, which is inefficient as it consumes a lot of workforce, time, and cost. Also, the accuracy of the manual sugarcane counting methods is not guaranteed because of the sugarcane field ergonomics and lack of skilled workers. The overhead of the manual counting methods motivated us to develop a novel Sugarcane Internode Detection and Counting Network (SIDCNet). Two versions of SIDCNet namely SIDCNet_Vl and SIDCNet_V2 are experimented in this work. This work collected a custom sugarcane internode dataset from Vellore sugarcane fields in Tamil Nadu, India, to improve the model performance. The proposed model utilises YOLOv4 as the base, where the pre-trained weights are retrained using both online and VIT's custom sugarcane internode dataset with 6682 images. The SIDCNet_ V2 model demonstrated a notable average accuracy of 95.37<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> in detecting and counting sugarcane internodes while exhibiting minimal false positives. The proposed model's performance was tested on real-time images and videos captured by a ground rover platform. Further, the inference results produced by the SIDCNets were compared with the ground truth values to ensure accuracy. The experimental results reveal that the proposed model can automate the sugarcane internode detection and counting process, which helps experts monitor the sugarcane internode growth and status throughout the crop period, thereby improving sugarcane productivity. In addition to the model experimentation on the custom sugarcane internode dataset, this research article also intends to reveal the methodologies and challenges of the custom sugarcane dataset collection process.