Single-class target detection method and device based on deep learning, and storage medium

Abstract

The invention provides a single-class target detection method and device based on deep learning and a storage medium. The method comprises the steps of obtaining door controller information and attention weight information corresponding to input through a trained depth multi-example change detection model based on two-period remote sensing image data by obtaining first-period sensing image data and second-period sensing image data of a to-be-detected region; according to the door controller information, obtaining scene-level change detection information of a target category object; and according to the attention weight information and the door controller information, using a contour extraction method to obtain a change instance graph of the target category object. The embodiment of the invention provides a method. Since the trained depth multi-instance change detection model is used to learn the depth features of the ground object from the remote sensing images in two periods, and themulti-instance learning framework is used to train the network through the scene-level annotation sample, a pixel-level annotation sample is not needed, the efficiency of single-class target detectionis improved, and the consumption of labor resources is reduced. 本发明提出了一种基于深度学习的单类目标检测方法、设备及存储介质，通过获取待测地区的第一时期和第二时期感影像数据，基于两时期遥感影像数据，通过已训练的深度多示例变化检测模型，得到与输入对应的门控制器信息和注意力权重信息；根据所述门控制器信息，得到目标类别物体的场景级变化检测信息；根据所述注意力权重信息和门控制器信息，使用轮廓提取方法，得到目标类别物体的变化实例图。本实施例提供的方法，由于使用已训练的深度多示例变化检测模型从两时期遥感影像中学习地物的深度特征，并使用多示例学习框架通过场景级标注样本对网络进行训练，无需要像素级标注样本，提高了单类目标检测的效率，减少了人工资源的消耗。