Grape multimodal image dataset

Current research on grape picking point detection mainly relies on grape detection and segmentation using visible RGB images. As we know, the accuracy of grape detection and segmentation is very helpful in improving the recognition result and localization accuracy of grape picking points. However, there are significant differences in background saliency among different varieties of grapes (green, purple), and their recognition rate and segmentation accuracy are usually rather poor in natural scenarios, because of leaf occlusion, overlapping fruits, complex background, and ever-changing illumination. Additionally, the grapes differ greatly from the morphological characteristics of other fruits such as apples and pears, making it more difficult for the existing dataset to meet the research requirements of the detection and segmentation of bunch-shaped grape fruits. To this end, the construction of visible, depth and infrared-based multimodal image datasets of grapes is crucial to explore higher recognition rates and stronger generalization capabilities for grape detection and semantic segmentation models. This dataset offers high-quality multimodal image data of two grape varieties with the colors of green (white Rose) and purple (Giant Peak) under different lighting and shading conditions. The number of images is 300 and 500 each for green and purple grapes, with a total of 3938 targets and approximately 2.97 GB. By means of rotation, deflation, mis-slicing, panning, and Gaussian blur, the dataset can be augmented for training implementation of mainstream deep learning models. The dataset can provide valuable basic image data resources for multimodal image data fusion, grape semantic segmentation, and object detection, which has important practical application value for promoting research in the field of agricultural machinery and equipment intelligence.