<p>PestMoth11 is a dual-domain image dataset designed to support image classification and object detection of 11 economically important agricultural pest moth species across controlled specimen and ecological imaging contexts. The dataset comprises 29,936 adult moth images from 11 species, including 20,078 multi-angle specimen images generated from 781 specimens, along with 9,858 ecological images sourced from iNaturalist observations that reflect field conditions with spatiotemporal occurrence metadata. All images have been meticulously annotated with species-level labels and YOLO (You Only Look Once) format bounding boxes. To validate the dataset’s utility, we conducted baseline experiments using representative architectures: EfficientNet-B0 and ResNet-50 for image classification, and YOLO11n and RT-DETR-l (Real-Time Detection Transformer-large) for object detection. Using stratified group-based splits defined by specimen identity and geographic grid grouping to reduce the risk of data leakage, baseline validation and multi-seed experiments produced classification macro F1-scores close to 0.99 and detection Mean Average Precision (mAP) at an Intersection over Union (IoU) thresholds ranging from 0.5 to 0.95 (mAP50–95) values close to 0.90 under the combined-domain setting, supporting the utility of the dataset for reproducible model evaluation. Cross-domain experiments revealed a clear domain gap between specimen-domain and ecological-domain images, whereas a cross-domain fusion strategy that combined both domains substantially improved performance compared with single-domain training. PestMoth11 supports dual-domain, dual-task moth pest recognition, contributing to the development of intelligent monitoring and early-warning systems in agricultural pest management. Furthermore, the dataset may facilitate multimodal pest monitoring studies that integrate image and metadata information.</p>