Image and Point Cloud-Based Neural Network Models and Applications in Agricultural Nursery Plant Protection Tasks

Abstract

Nurseries represent a fundamental component of modern agricultural systems, specializing in the cultivation and management of diverse seedlings. Scientific cultivation methods significantly enhance seedling survival rates, while intelligent agricultural robots improve operational efficiency through autonomous plant protection. Central to these robotic systems, the perception system utilizes advanced neural networks to process environmental data from both images and point clouds, enabling precise feature extraction. This review systematically explores prevalent image-based models for classification, segmentation, and object detection tasks, alongside point cloud processing techniques employing multi-view, voxel-based, and original data approaches. The discussion extends to practical applications across six critical plant protection areas. Image-based neural network models can fully utilize the color information of objects, making them more suitable for tasks such as leaf disease detection and pest detection. In contrast, point cloud-based neural network models can take full advantage of the spatial information of objects, thus being more applicable to tasks like target information detection. By identifying current challenges and future research priorities, the analysis provides valuable insights for advancing agricultural robotics and precision plant protection technologies.