Multitask learning is a widely recognized technique in the field of computer vision and deep learning domain. However, it is still a research question in remote sensing, particularly for hyperspectral imaging (HSI). Moreover, most of the research in the remote sensing domain focuses on small and single-task-based annotated datasets, which limits the generalizability and scalability of the developed models to more diverse and complex real-world scenarios. Thus, in this study, we propose a multitask deep learning model designed to perform multiple classification and regression tasks simultaneously on hyperspectral images. We validated our approach on a large hyperspectral dataset called TAIGA, which contains 13 forest variables, including three categorical variables and ten continuous variables with different biophysical parameters. We design a sharing encoder and task-specific decoder network to streamline feature learning while allowing each task-specific decoder to focus on the unique aspects of its respective task. In addition, a dense atrous pyramid pooling layer and attention network were integrated to extract multiscale contextual information and enable selective information processing by prioritizing task-specific features. Furthermore, we computed multitask loss and optimized its parameters for the proposed framework to improve the model performance and efficiency across diverse tasks. A comprehensive qualitative and quantitative analysis of the results shows that the proposed method significantly outperforms other state-of-the-art methods. We trained our model across ten seeds/trials to ensure robustness. Our proposed model demonstrates higher mean performance while maintaining lower or equivalent variability. To make the work reproducible, the codes will be available athttps://github.com/Koushikey4596/Multitask-Deep-Learning-Model-for-Taiga-datatset.

Hyperspectral imaging is essential for the detailed and accurate identification of materials and features across various applications, including environmental monitoring and agricultural assessment. However, processing large-scale hyperspectral images is time-consuming and requires huge computational resources attributed to their volume. Additionally, classification methods lacking spatial information and focusing only on spectral information are inadequate with the feature complexity of the dataset. Hence, to overcome the above-mentioned challenges, there is a dire need for advanced algorithms capable of efficiently processing large-scale hyperspectral data. This paper highlights the motivation and need for multitask learning models for hyperspectral datasets. A large-scale hyperspectral dataset called TAIGA, which comprises both categorical and continuous forest variables, is considered. The proposed deep learning model addresses data imbalance and loss function concerns by incorporating spectral as well as spatial information to develop a multitask model. This work emphasizes the importance of accounting for correlations between data and tasks when designing a relevant model. As a result, the study leads to more efficient model training, reduced data requirements, and potentially improved overall predictive capabilities. We achieved overall accuracy for categorical variables as high as 98.25% and mean absolute error as low as 0.019 for continuous variables.

Recently, Convolutional Neural Network (CNN) has been widely used for the classification of hyperspectral images. However, with the availability of limited training sample data in hyperspectral scenes, classification performance is majorly affected. In this research work, we propose a shuffled dilated CNN-based classification method for hyperspectral classification using a transfer learning approach. The proposed model consists of a dilated convolution layer to provide a larger receptive field and a shuffled block to enhance the connection between different layers. This helps in the development of a classification model having fewer parameters with better efficiency. To evaluate the performance, hyperspectral datasets have been considered from the same sensor as well as cross-sensor with similar spectral and spatial features. We observed an improvement of 4% overall accuracy using the transfer learning approach compared to the classification results without using the transfer learning approach. The experimental results demonstrate the effectiveness of using the proposed methodology with a transfer learning approach for the classification of hyperspectral data with a limited number of labeled training samples.

Salient object detection using hyperspectral images is crucial for various image processing and computer vision applications. Many studies considering spectral information have been developed, extracting only low-level features from a hy-perspectral image. In this research work, a dataset specifically developed for salient object detection called HS-SOD is considered exploiting both spatial and spectral information equally. To include spatial information, Extended Morpho-logical Profile (EMP) has been considered. EMP incorpo-rates spatial characteristics by including nearby pixel information. A convolution neural network (CNN) is integrated with extended morphology to extract high-level features. It detect objects of multiple spatial scales and ratios, preserving boundary edges. We observed an improvement of 5 % in overall accuracy while using EMP with CNN compared to that of using EMP without CNN. Thus, the experimental re-sults demonstrate the effectiveness of EMP with CNN on the hyperspectral datasets.

Salient object detection in hyperspectral images is of interest in various image processing and computer vision applications. Many studies considering spectral information have been reported, extracting only low-level features from a hyperspectral image. This paper proposes a Convolutional Neural Network (CNN) based salient object detection method using hyperspectral imagery to utilise spatial and spectral information simultaneously. The proposed methodology incorporates Extended Morphological Profile (EMP) followed by a CNN to utilise the information from nearby pixels and high-level features simultaneously. We have evaluated the performance of the proposed approach on two independent datasets to verify the generalisation ability, viz. 1) Hyperspectral Salient Object Detection Dataset (HS-SOD) and 2) Pavia University dataset. An extensive quantitative analysis of the results revealed that the proposed method significantly outperforms other state-of-the-art methods by approximately ≥ 2% of AUC (Area Under receiver operating characteristic Curve) and F-measure and lower mean absolute error for both datasets.