Semantic Segmentation

This paper dives into the cutting-edge world of road asset detection on Thai highways, showcasing a novel approach that combines an upgraded REG model with Generalized Focal Loss. Our focus is on identifying key road elements—like pavilions, pedestrian bridges, information and warning signs, and concrete guardrails—to boost road safety and infrastructure management. While deep learning methods have shown promise, traditional models often struggle with accuracy in tricky conditions, such as cluttered backgrounds and variable lighting. To tackle these issues, we’ve integrated REG with Generalized Focal Loss, enhancing its ability to detect road assets with greater precision. Our results are impressive, the REGx model led the way with a mAP50 of 80.340, mAP50-95 of 60.840, precision of 79.100, recall of 76.680, and an F1-score of 77.870. These findings highlight the REGx model’s superior performance, demonstrating the power of advanced deep learning techniques to improve highway safety and infrastructure maintenance, even in challenging conditions.

In this paper, we present MeViT (Medium-Resolution Vision Transformer), designed for semantic segmentation of Landsat satellite imagery, focusing on key economic crops in Thailand para rubber, corn, and pineapple. MeViT enhances Vision Transformers (ViTs) by integrating medium-resolution multi-branch architectures and revising mixed-scale convolutional feedforward networks (MixCFN) to extract multi-scale local information. Extensive experiments on a public Thailand dataset demonstrate that MeViT outperforms state-of-the-art deep learning methods, achieving a precision of 92.22%, recall of 94.69%, F1 score of 93.44%, and mean IoU of 83.63%. These results highlight MeViT’s effectiveness in accurately segmenting Thai Landsat-8 data.

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.

My PhD thesis focuses on improving semantic segmentation of aerial and satellite images, a crucial task for applications like agriculture planning, map updates, route optimization, and navigation. Current models like the Deep Convolutional Encoder-Decoder (DCED) have limitations in accuracy due to their inability to recover low-level features and the scarcity of training data. To address these issues, I propose a new architecture with five key enhancements, a Global Convolutional Network (GCN) for improved feature extraction, channel attention for selecting discriminative features, domain-specific transfer learning to address data scarcity, Feature Fusion (FF) for capturing low-level details, and Depthwise Atrous Convolution (DA) for refining features. Experiments on Landsat-8 datasets and the ISPRS Vaihingen benchmark showed that my proposed architecture significantly outperforms the baseline models in remote sensing imagery.

For an urban development, the Quality of Life (QOL) of people in the city is a vital issue that should be considered. There are many researches in QOL topics that use questionnaire survey approach. These studies yield very useful information for city development planning. As the Artificial Intelligence technologies are developed very fast recently, they are applied to solve many transportation problems. In this paper, we propose a method that automatically extract mobility indicators using two image recognition techniques, Semantic Segmentation and Object Recognition. Because the mobility is an important factor in QOL evaluation, our work can be used to enhance a performance and reduce a data gathering cost of the QOL evaluation.

Semantic Segmentation is a fundamental task in computer vision and remote sensing imagery. Many applications, such as urban planning, change detection, and environmental monitoring, require the accurate segmentation; hence, most segmentation tasks are performed by humans. Currently, with the growth of Deep Convolutional Neural Network (DCNN), there are many works aiming to find the best network architecture fitting for this task. However, all of the studies are based on very-high resolution satellite images, and surprisingly; none of them are implemented on medium resolution satellite images. Moreover, no research has applied geoinformatics knowledge. Therefore, we purpose to compare the semantic segmentation models, which are FCN, SegNet, and GSN using medium resolution images from Landsat-8 satellite. In addition, we propose a modified SegNet model that can be used with remote sensing derived indices. The results show that the model that achieves the highest accuracy RGB bands of medium resolution aerial imagery is SegNet. The overall accuracy of the model increases when includes Near Infrared (NIR) and Short-Wave Infrared (SWIR) band. The results showed that our proposed method (our modified SegNet model, named RGB-IR-IDX-MSN method) outperforms all of the baselines in terms of mean F1 scores.

Semantic segmentation of remotely-sensed aerial (or very-high resolution, VHS) images and satellite (or high-resolution, HR) images has numerous application domains, particularly in road extraction, where the segmented objects serve as essential layers in geospatial databases. Despite several efforts to use deep convolutional neural networks (DCNNs) for road extraction from remote sensing images, accuracy remains a challenge. This paper introduces an enhanced DCNN framework specifically designed for road extraction from remote sensing images by incorporating landscape metrics (LMs) and conditional random fields (CRFs). Our framework employs the exponential linear unit (ELU) activation function to improve the DCNN, leading to a higher quantity and more accurate road extraction. Additionally, to minimize false classifications of road objects, we propose a solution based on the integration of LMs. To further refine the extracted roads, a CRF method is incorporated into our framework. Experiments conducted on Massachusetts road aerial imagery and Thailand Earth Observation System (THEOS) satellite imagery datasets demonstrated that our proposed framework outperforms SegNet, a state-of-the-art object segmentation technique, in most cases regarding precision, recall, and F1 score across various types of remote sensing imagery.

In this paper, we introduce an improved deep convolutional encoder-decoder network (DCED) for segmenting road objects from aerial images. Enhancements include the use of ELU (exponential linear unit) instead of ReLU, dataset augmentation with incrementally-rotated images to increase training data by eight times, and the use of landscape metrics to remove false road objects. Tested on the Massachusetts Roads dataset, our method outperformed the SegNet benchmark and other baselines in precision, recall, and F1 scores.