Search

Scene classification is a crucial research problem in remote sensing (RS) that has attracted many researchers recently. It has many challenges due to multiple issues, such as: the complexity of remote sensing scenes, the classes overlapping (as a scene may contain objects that belong to foreign classes), and the difficulty of gaining sufficient labeled scenes. Deep learning (DL) solutions and in particular convolutional neural networks (CNN) are now state-of-the-art solution in RS scene classification; however, CNN models need huge amounts of annotated data, which can be costly and time-consuming. On the other hand, it is relatively easy to acquire large amounts of unlabeled images. Recently, Self-Supervised Learning (SSL) is proposed as a method that can learn from unlabeled images, potentially reducing the need for labeling. In this work, we propose a deep SSL method, called RS-FewShotSSL, for RS scene classification under the few shot scenario when we only have a few (less than 20) labeled scenes per class. Under this scenario, typical DL solutions that fine-tune CNN models, pre-trained on the ImageNet dataset, fail dramatically. In the SSL paradigm, a DL model is pre-trained from scratch during the pretext task using the large amounts of unlabeled scenes. Then, during the main or the so-called downstream task, the model is fine-tuned on the labeled scenes. Our proposed RS-FewShotSSL solution is composed of an online network and a target network both using the EfficientNet-B3 CNN model as a feature encoder backbone. During the pretext task, RS-FewShotSSL learns discriminative features from the unlabeled images using cross-view contrastive learning. Different views are generated from each image using geometric transformations and passed to the online and target networks. Then, the whole model is optimized by minimizing the cross-view distance between the online and target networks. To address the problem of limited computation resources available to us, our proposed method uses a novel DL architecture that can be trained using both high-resolution and low-resolution images. During the pretext task, RS-FewShotSSL is trained using low-resolution images, thereby, allowing for larger batch sizes which significantly boosts the performance of the proposed pipeline on the task of RS classification. In the downstream task, the target network is discarded, and the online network is fine-tuned using the few labeled shots or scenes. Here, we use smaller batches of both high-resolution and low-resolution images. This architecture allows RS-FewshotSSL to benefit from both large batch sizes and full image sizes, thereby learning from the large amounts of unlabeled data in an effective way. We tested RS-FewShotSSL on three RS public datasets, and it demonstrated a significant improvement compared to other state-of-the-art methods such as: SimCLR, MoCo, BYOL and IDSSL. [ABSTRACT FROM AUTHOR]

Advances in technology can provide a lot of support for visually impaired (VI) persons. In particular, computer vision and machine learning can provide solutions for object detection and recognition. In this work, we propose a multi-label image classification solution for assisting a VI person in recognizing the presence of multiple objects in a scene. The solution is based on the fusion of two deep CNN models using the induced ordered weighted averaging (OWA) approach. Namely, in this work, we fuse the outputs of two pre-trained CNN models, VGG16 and SqueezeNet. To use the induced OWA approach, we need to estimate a confidence measure in the outputs of the two CNN base models. To this end, we propose the residual error between the predicted output and the true output as a measure of confidence. We estimate this residual error using another dedicated CNN model that is trained on the residual errors computed from the main CNN models. Then, the OAW technique uses these estimated residual errors as confidence measures and fuses the decisions of the two main CNN models. When tested on four image datasets of indoor environments from two separate locations, the proposed novel method improves the detection accuracy compared to both base CNN models. The results are also significantly better than state-of-the-art methods reported in the literature. [ABSTRACT FROM AUTHOR]

During the past few years, cluster ensemble methods have been introduced in the area of pattern recognition as a more accurate alternative to individual clustering algorithms as they can provide novel, robust, and stable solutions. The development of these methods is mainly motivated by the success of the combination of supervised classifiers also known as classifier ensemble. In this paper, we propose a novel ensemble method for the classification of hyperspectral images based on Fuzzy c-means (FCM) and Markov random field (MRF) theory. Firstly, in order to construct the ensemble of clustering maps, which call here 'partitions', we run the FCM algorithm several times with different initializations and different subsets of features (spectral bands) selected randomly from the high-dimensional input feature vector characterizing the hyperspectral image. Secondly, since there is no-ground truth information, the class labels contained in each generated partition are symbolic, meaning that the same class of pixels can be labeled with different labels in different partitions of the ensemble. Therefore, an optimal relabeling of the ensemble with respect to a representative partition determined on the basis of the maximum entropy principle is made via a pairwise relabeling procedure. Finally, in the last step, the relabeled partitions are fused by means of an MRF method. The latter exploits two kinds of sources of contextual information: spatial and inter-partition contextual information. During the optimization process, the contributions of the different partitions are controlled through mutual information weights. [ABSTRACT FROM AUTHOR]