2026-10-12T12:45:54Z https://www.scidb.cn/oai

10.57760/sciencedb.46312 2026-08-14T10:15:45Z

2026-08-14 Multimodal Visual Instruction Dataset for Intelligent Transmission Line Inspection: Single-Turn Fine-Grained Description Edition doi:10.57760/sciencedb.46312 en This dataset was developed for intelligent transmission-line inspection and instruction tuning of multimodal vision-language models. It contains representative inspection scenes involving transmission lines, transmission towers, insulators, conductors, and foreign objects in power-line corridors. The single-turn fine-grained description edition contains 9,445 image-instruction samples, including 7,599 training samples and 1,846 validation samples. The training split consists of 5,765 original samples retained after duplicate and data-leakage screening and 1,834 class-aware augmented samples.The dataset covers balloons, bird nests, kites, small transmission towers, transmission towers, insulators, broken insulators, flashed insulators, insulators in good condition, damaged conductor insulation, broken conductor strands, and loose conductor strands. Each sample contains one inspection image, one user instruction, and one detailed Chinese response. The response is written according to the actual visible content and describes the inspection environment, equipment type and structure, anomaly location, appearance characteristics, defect condition, potential operational risks, and recommended maintenance actions.The dataset follows a conversational instruction format suitable for fine-tuning multimodal large language models such as Qwen-VL. Images are referenced using relative paths to facilitate data migration, model training, and experiment reproduction. Before release, the dataset was checked for image-annotation pairing, file readability, duplicate records, train-validation leakage, category validity, semantic coverage, conversation structure, and path portability. The responses do not contain object-detection procedures, region coordinates, or bounding-box descriptions. The dataset can support research on transmission-line image captioning, visual question answering, defect recognition, condition assessment, risk analysis, multimodal instruction tuning, and intelligent inspection report generation. transmission line inspection; multimodal vision-language model; visual instruction tuning; power equipment defects; image captioning; visual question answering; insulator defects; conductor defects Xie Xiaowei Cao Hui Dou Liyang Zhang Jingxuan PUBLIC https://creativecommons.org/licenses/by-nc/4.0/ dataset Science Data Bank