Abstract: Fruit-tree canopies are typically characterized by complex branching architecture and dense foliage, which causes severe self-occlusion, uneven light distribution, and consequently low light-use efficiency. In production, pruning decisions are still largely experience-driven and may be inconsistent across operators and orchards, leading to non-standard canopy shapes and suboptimal regulation of tree vigor. To address these challenges, this study proposes an intelligent pruning recommendation model that jointly optimizes canopy structure reconstruction and light-efficiency evaluation based on Neural Radiance Fields (NeRF). The goal is to provide a reproducible, quantitative, and visualization-friendly workflow for canopy analysis and pruning guidance under real orchard conditions. Qingcuili plum (Prunus salicina cv. ‘Qingcuili’) was selected as the target species. Multi-view videos were captured around each tree from multiple angles to ensure sufficient coverage of the canopy under field lighting and background conditions. A NeRF-based reconstruction pipeline was used to learn volumetric radiance and density fields from the video frames, generating a high-fidelity 3D representation of the tree. From the reconstructed structure, branch topology and geometric descriptors (e.g., branch order, orientation, length, and spatial distribution) were extracted to support subsequent decision-making. To quantify light conditions, a Monte Carlo Ray Tracing (MCRT) module was integrated to simulate ray–canopy interactions and to estimate both direct and diffuse radiation pathways in the 3D canopy space. Two key indicators were computed: light interception ratio (LIR) to characterize the proportion of incident light intercepted by the canopy, and energy interception ratio (EIR) to reflect the effective energy captured under the simulated radiation field. On this basis, a pruning recommendation model was constructed by fusing (i) branch-classification and geometric features and (ii) light-efficiency weights derived from the MCRT outputs, enabling ranked candidate-branch suggestions that target improved canopy illumination while maintaining structural feasibility. In addition, a virtual interactive pruning system was developed to support human-in-the-loop validation, intuitive visualization of light distribution, and examination of pruning effects before field implementation. The proposed framework was validated on a dataset of 102 Qingcuili plum trees with diverse canopy forms and growth conditions. The NeRF-based reconstruction achieved high accuracy, with an average reconstruction error of 4.3% ± 1.6%, indicating reliable recovery of canopy structure under complex occlusion. The pruning recommendation performance reached 91% accuracy when compared with expert labelling, demonstrating that the model can provide pruning suggestions consistent with practical management knowledge. After applying the recommended pruning strategy in simulation, canopy light conditions improved substantially: LIR increased by approximately 15%, and EIR increased by nearly 19%, reflecting enhanced light interception and more effective energy capture. From a deployment perspective, with cloud-based GPU acceleration (NVIDIA V100), the end-to-end system response time remained stable within 2.3 minutes, suggesting favourable real-time applicability for orchard decision support and interactive analysis. This study presents a canopy structure–light-efficiency collaborative optimization approach that integrates NeRF-based 3D reconstruction, MCRT-based light simulation, and feature–weight fusion for intelligent pruning recommendation. The proposed model enables practical 3D canopy reconstruction and quantitative pruning guidance in real orchard environments, improves light-use efficiency and the precision of tree vigour regulation, and provides a feasible technical pathway toward digital twins and smart pruning for fruit-tree production.