Three-dimensional Reconstruction and Picking Point Localization Method of Cucumber Fruit Based on Point Cloud Completion

In practical cucumber harvesting scenarios, the presence of direct sunlight and occlusion in greenhouses poses significant challenges for depth cameras to provide sufficient three-dimensional environmental information for harvesting robots. To address this issue, this study innovatively applies point cloud completion techniques to picking point localization, proposing a novel method for cucumber picking point localization based on a point cloud completion network. This method aligns the completed point cloud with the two-dimensional coordinates of picking points identified in RGB images, enabling precise acquisition of the 3D coordinates and pose information of the picking points. A new point cloud completion network is introduced, incorporating a Dual Attention-based Feature Extraction Module (DA-FEM) that weights and filters features to effectively capture local information while suppressing interference, thereby reducing noise. Additionally, to overcome the limitations of traditional fusion methods that restrict the model’s receptive field, a Multi-Scale Hierarchical Feature Fusion Module (MS-HFFM) is proposed, complemented by a pyramid-structured point cloud generator to enhance detail recovery. Furthermore, a joint loss function based on Chamfer Distance (CD) and Earth Mover Distance (EMD) is designed to minimize stitching artifacts in the completed point cloud, thereby improving its quality. Experimental results demonstrate that the proposed model achieves average CD and EMD values of 0.82×10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">–3</sup> and 2.712×10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">–3</sup>, respectively, on an augmented ShapeNetCore dataset, with the picking point localization method attaining an average accuracy of 90.40%. These findings provide robust support for the development of cucumber harvesting robots.