Canopy segmentation in orchards is crucial for obtaining vegetation biometric parameters, especially with high-resolution images from UAV. Current methods include Object-Based Image Analysis (OBIA) and deep learning (DL) techniques. This study compares the performance of the U-Net method with an OBIA method, the Gaussian Mixture Model (GMM), and the K-Means clustering algorithm. UAV-based multispectral RGB-NIR images were captured in a vineyard in northwest Sicily. After extracting pure canopy pixels, the NDVI index was calculated and compared with canopy agronomic data. Parameters such as leaf area (LA) and leaf chlorophyll content (LCC) were considered. The results show that the U-Net model outperforms OBIA and K-Means in extracting canopy pixels, with an accuracy of 90%, an F1_score of 88%, and an mIoU of 80%. The accuracy of the U-Net method justifies the consistent correlations between the NDVI index and canopy parameters such as LA and LCC, with determination coefficients of (r = 0.85) and (r = 0.88). Segmentation and classification methods significantly influence the estimation of canopy biometric parameters.