Vision Transformer

Semantic segmentation on Landsat-8 data is crucial in the integration of diverse data, allowing researchers to achieve more productivity and lower expenses. This research aimed to improve the versatile backbone for dense prediction without convolutions—namely, using the pyramid vision transformer (PRM-VS-TM) to incorporate attention mechanisms across various feature maps. Furthermore, the PRM-VS-TM constructs an end-to-end object detection system without convolutions and uses handcrafted components, such as dense anchors and non-maximum suspension (NMS). The present study was conducted on a private dataset, i.e., the Thailand Landsat-8 challenge. There are three baselines, DeepLab, Swin Transformer (Swin TF), and PRM-VS-TM. Results indicate that the proposed model significantly outperforms all current baselines on the Thailand Landsat-8 corpus, providing F1-scores greater than 80% in almost all categories. Finally, we demonstrate that our model, without utilizing pre-trained settings or any further post-processing, can outperform current state-of-the-art (SOTA) methods for both agriculture and forest classes.

Transformers have demonstrated remarkable accomplishments in several natural language processing (NLP) tasks as well as image processing tasks. Herein, we present a deep-learning (DL) model that is capable of improving the semantic segmentation network in two ways. First, utilizing the pre-training Swin Transformer (SwinTF) under Vision Transformer (ViT) as a backbone, the model weights downstream tasks by joining task layers upon the pretrained encoder. Secondly, decoder designs are applied to our DL network with three decoder designs, U-Net, pyramid scene parsing (PSP) network, and feature pyramid network (FPN), to perform pixel-level segmentation. The results are compared with other image labeling state of the art (SOTA) methods, such as global convolutional network (GCN) and ViT. Extensive experiments show that our Swin Transformer (SwinTF) with decoder designs reached a new state of the art on the Thailand Isan Landsat-8 corpus (89.8% 𝐹1 score), Thailand North Landsat-8 corpus (63.12% 𝐹1 score), and competitive results on ISPRS Vaihingen. Moreover, both our best-proposed methods (SwinTF-PSP and SwinTF-FPN) even outperformed SwinTF with supervised pre-training ViT on the ImageNet-1K in the Thailand, Landsat-8, and ISPRS Vaihingen corpora.