Scholarlink

In the realm of Artificial Intelligence of Everything (AIoE), remote sensing finds application through the uti-lization of observations and data collection, primarily in the form of images, obtained from a distance, often via satellites or other aerial platforms. The processing and extraction of meaningful insights from these remote sensing images (RSIs) present notable challenges, owing to the vast amount of data and the presence of noise and artifacts. To tackle this, there emerges a pressing need for AI models adept at interpreting RSIs and extracting relevant information, catering to both novice users and expert scientists in the field. This chapter delves into the domain of Remote Sensing Visual Question Answering (RSVQA), a subset of AI models rooted in computer vision, tailored to respond to natural language queries about RSIs. While still nascent, RSVQA models, initially pioneered around 2020, hold substantial potential for the development of RSI-centric AI search engines. The chapter offers a comprehensive examination of existing RSVQA techniques, exploring their underlying principles, practical applications, and inherent constraints, while also introducing an innovative RSVQA model. The proposed model comprises five interconnected modules: Input, Pre-processing, Representation, Fusion, and Answering. The Input module receives RSIs captured by various sensors alongside relevant queries in the form of natural language questions. The Pre-processing module incorporates a Denoising sub-module along with an natural language pre-processing sub-module, aimed at refining the input data and preparing it for subsequent processing. The Representation module extracts distinctive features from the inputs via visual and question representation sub-modules, which are then consolidated into a vectorized encoding through the Fusion module. Finally, the Answering module employs a multilayer perceptron (MLP) classifier to decipher the desired output. A comparative analysis enriches the chapter, offering valuable insights for practitioners involved in RSVQA research and development. The chapter culminates by contemplating the future trajectory of RSVQA, taking into account emerging trends, ethical considerations, and potential industry implications, thereby providing a forward-looking perspective on this evolving facet of AIoE.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Remote Sensing in the Era of Artificial Intelligence of Everything Through Visual Question Answering

  • 章节

  • 发表日期：

    2025年

查看全文
  • Anirban Saha,
  • Suman Kumar Maji

摘要

In the realm of Artificial Intelligence of Everything (AIoE), remote sensing finds application through the uti-lization of observations and data collection, primarily in the form of images, obtained from a distance, often via satellites or other aerial platforms. The processing and extraction of meaningful insights from these remote sensing images (RSIs) present notable challenges, owing to the vast amount of data and the presence of noise and artifacts. To tackle this, there emerges a pressing need for AI models adept at interpreting RSIs and extracting relevant information, catering to both novice users and expert scientists in the field. This chapter delves into the domain of Remote Sensing Visual Question Answering (RSVQA), a subset of AI models rooted in computer vision, tailored to respond to natural language queries about RSIs. While still nascent, RSVQA models, initially pioneered around 2020, hold substantial potential for the development of RSI-centric AI search engines. The chapter offers a comprehensive examination of existing RSVQA techniques, exploring their underlying principles, practical applications, and inherent constraints, while also introducing an innovative RSVQA model. The proposed model comprises five interconnected modules: Input, Pre-processing, Representation, Fusion, and Answering. The Input module receives RSIs captured by various sensors alongside relevant queries in the form of natural language questions. The Pre-processing module incorporates a Denoising sub-module along with an natural language pre-processing sub-module, aimed at refining the input data and preparing it for subsequent processing. The Representation module extracts distinctive features from the inputs via visual and question representation sub-modules, which are then consolidated into a vectorized encoding through the Fusion module. Finally, the Answering module employs a multilayer perceptron (MLP) classifier to decipher the desired output. A comparative analysis enriches the chapter, offering valuable insights for practitioners involved in RSVQA research and development. The chapter culminates by contemplating the future trajectory of RSVQA, taking into account emerging trends, ethical considerations, and potential industry implications, thereby providing a forward-looking perspective on this evolving facet of AIoE.