Sweet Potato Instance Segmentation Dataset

The current agriculture industry faces numerous challenges in crop monitoring and pest detection, particularly in the application of image segmentation technology, where traditional methods often fail to meet practical needs. Existing solutions largely rely on manual identification, which is inefficient and lacks precision. This dataset aims to provide a high-quality sweet potato instance segmentation dataset to support the training and research of deep learning models. The dataset contains 5000 sweet potato images collected from various planting environments and captured using high-definition cameras to ensure image clarity and detail. During data collection, multiple rounds of annotation were carried out, with consistency checks and reviews conducted by professionals to ensure data quality. The data is stored in PNG and JPG formats to facilitate subsequent processing and analysis. The core advantage of this dataset lies in its high annotation accuracy and consistency, with an annotation accuracy reaching 95% through expert review. Advanced data augmentation techniques enhance the model's generalization ability, solving the poor performance of traditional methods in complex environments. Models trained using this dataset showed a 20% improvement in accuracy for pest detection, providing strong support for agricultural intelligence.