Accurate under-canopy weed detection is a major challenge in precision agriculture due to occlusion in conventional UAV imagery. This study presents an unsupervised 3D point-cloud framework integrating UAV and nano-drone data to improve weed identification in dense crop environments. Field experiments in a blueberry plantation combined UAV (DJI Mavic 3M) and nano-drone (DJI Mini 4 Pro) imagery. Point clouds were co-registered using ground control points, rigid transformation, and Iterative Closest Point (ICP). Segmentation employed Progressive Morphological Filtering (PMF) for ground/non-ground separation, weighted K-means for soil–vegetation classification, and DBSCAN–K-means clustering for individual weed identification. A 2D weed map was generated by projecting crops and weed location. The integrated dataset identified 125 weed clusters compared to 52 with UAV-only data, achieving 72.8% precision, 93.8% recall, and 82.1% F1-score. Results confirm that multi-perspective point clouds enhance under-canopy weed detection and provide a foundation for large-scale weed mapping, and automated weeding systems.