A Review and a Perspective of Deep Active Learning for Remote Sensing Image Analysis: Enhanced adaptation to user conjecture

In recent years, the application of deep learning (DL) has revolutionized remote sensing (RS) image analysis, allowing for the extraction of high-level features, and addressing complex tasks. However, the success of DL models relies heavily on the availability of labeled data, and acquiring labeled samples in the RS domain can be particularly challenging due to factors such as cost, time, and the dynamic nature of landscapes. Active learning (AL) has been a well-established concept in RS imagery analysis, even predating the widespread adoption of DL. Its significance lies in its ability to iteratively select the most informative samples from the unlabeled dataset, reducing the annotation cost and improving model performance with limited labeled data. In the era of DL, where the demand for labeled samples is higher than ever, AL has become increasingly crucial. Deep AL is an innovative and intricate approach that seeks to harness the strengths of both DL and AL methodologies. This integration aims to improve the performance of DL models while reducing the reliance on large amounts of labeled data. Integrating AL with deep architectures presents challenges but offers a promising approach to RS tasks.