Search

Existing hyperspectral image (HSI) classification methods rarely consider open-set classification (OSC). Although some reconstruction-based methods can deal with OSC, they lack adaptive threshold strategies and heavily rely on the labeled samples. Therefore, this article proposes a self-supervised multitask learning (SSMTL) framework for few-shot open-set HSI classification, including three stages: pretraining stage (PTS), fine-tuning stage, and testing stage. The model consists of three modules: data diversification module (DDM), 3-D multiscale attention module (3D-MAM), and adaptive threshold module (ATM), as well as a backbone network: dense feature pyramid network (DFPN). In the PTS, we construct a self-supervised reconstruction task with unlabeled samples for model initialization, where DDM aims to improve the robustness of the model and 3D-MAM applies 3-D multiscale convolution to focus on key information spatially and spectrally. In the fine-tuning stage, we further optimize the model with a few labeled samples based on both reconstruction task and classification task, where ATM implements adaptive threshold strategies based on uncertainties of predicted probability and reconstruction loss, and DFPN is helpful to retain the detailed information. The experimental results on three common HSI datasets show SSMTL performs significantly well and even surpasses many advanced closed-set and open-set HSI classification methods.

Precipitable water vapor (PWV), as an important representative parameter of atmospheric water vapor contents, can be obtained by means of Global Navigation Satellite Systems (GNSS) using both ground-based and space-borne observation techniques. However, the PWV prediction models currently accessible tend to be simplistic combinations or individual models. In this study, we develop a whale optimization algorithm (WOA) convolutional neural network (CNN) long short-term memory (LSTM)-Attention model to predict PWV, which takes the 16 GNSS PWV values near the King’s Park (HKKP) station as characteristic parameters and the spatial relationship between the point of interest and its neighboring GNSS stations into consideration. An optimal model via the WOA is investigated by using a wavelet analysis to separate noises, through combining CNN, LSTM neural network, and attention mechanism. Results show that considerable improvement in the prediction accuracy has been achieved through a comparison between CNN-LSTM–Attention and the conventional LSTM and CNN-LSTM models. In terms of long-term predictability, CNN-LSTM–Attention is proven to be a superior model when eight features are incorporated. The model’s root mean square error (RMSE) is 2.30 mm, which is reduced by 20.42% than in the case of 0 feature is used. As a further analysis, we also examine the prediction performance of various models for hourly PWV using 7, 15, 30, 60, and 90 days of data as different lengths of training. The results show that CNN-LSTM–Attention has a better prediction effect when the training length is 30 days, the RMSE is 0.74 mm, and the Nash-Sutcliffe efficiency (NSE) coefficient is 0.98.