Work-in-Progress Paper: Synergizing YOLOv8 and PCA for Size Estimation of Baby Broccoli

This research aims to advance agricultural automation by developing a machine learning-based method for accurately measuring harvest-ready baby broccoli through estimating the size of individual heads. Unlike traditional broccoli, baby broccoli’s varying head depth complicates size estimation, making direct pixel-size conversion ineffective. To overcome this, we utilized depth-sensing cameras to capture precise dimensions. Our initial efforts involved curating a unique dataset and applying the YOLO computer vision algorithm for segmenting baby broccoli heads. We then calculated the size of each identified head using Principal Component Analysis (PCA). Given that baby broccoli tends to grow in tight clusters, our method includes tracking individual heads within a frame and associating them with specific size information to ensure accurate management of information. By incorporating stereo vision and depth data from the realsense D435 camera along with instance segmentation and PCA, our initial results achieved size estimates with an error rate below $10 \%$. The paper further recommends enhancing accuracy through a hybrid approach that combines deep learning, neural networks, and PCA.