DYNAMIC COLORIZATION OF SAR IMAGES POWERED BY LARGE LANGUAGE MODEL INTERFACE

Abstract

Dynamic colorization of SAR images powered by large language model interface combines Conditional GANs (cGANs) and ResNet for SAR image colorization and enhanced by Generative Adversarial Refinement (GAR) for real-time user-driven adjustments. Users can interactively refine colorization through queries, leveraging LlaMA to interpret and adjust specific image attributes like water or vegetation. The iterative feedback loop between users and the GAR model enables dynamic and personalized image enhancements. This level of interactivity, combined with deep learning-based colorization, sets it apart from traditional SAR colorization methods, offering both accuracy and customization. LlaMA allows users to query specific details and receive real-time responses, which enhances the overall utility of the system.

Specification

Description:The system utilizes a Conditional Generative Adversarial Network (cGAN) as its core colorization engine. This model learns the mapping between grayscale SAR images and colorized outputs, enabling the generation of visually realistic images. The ResNet architecture, integrated within the cGAN, ensures that the model retains critical structural details in the images while applying color. Unlike traditional GANs, this system incorporates Generative Adversarial Refinement (GAR), allowing for real-time feedback-driven refinement. Users can query specific attributes of the image, such as "brighten water areas" or "adjust the green hue for vegetation." The GAR model processes these inputs and applies appropriate modifications to the image, leading to enhanced customization and accuracy. , C , Claims:1) Dynamic colorization of SAR images powered by large language model (101) interface uses the SAR image as input and make colorized image using cGAN and allows user to customize the image coloring using large language model (101) comprises:
i) Takes the SAR image (102) that needs to be analyzed and colorized;
ii) Given input image is pre-processed and make it ready for the feature extraction. ResNet model (103) is used to extract meaningful features from the SAR image (102);
iii) cGAN (104) takes the processed input image and the extracted features (103) to color the image;
iv) User can customize the color (106) for a specific location on the image using by giving instructions to LLM model (105);
v) LLM model converts the user requirement into appropriate instructions to cGAN model (104);
vi) Final version of the colorized image is used as input for LLM model (105) for identifying the textual insights (108);
vii) The image and the textual insights are stored in a database for further processing if needed;
2) Dynamic colorization of SAR images powered by large language model (101) interface uses pre-processing techniques on the input SAR image (102) to make it ready for feature extraction as claimed in claim 1.
3) Dynamic colorization of SAR images powered by large language model (101) interface uses the ResNet model (103) to extract the meaningful features for coloring from the SAR image (102) as claimed in claim 1.
4) Dynamic colorization of SAR images powered by large language model (101) interface uses the cGAN (104) to colorize the SAR image (102) based on the extracted features (103) from the ResNet model as claimed in claim 1.
5) Dynamic colorization of SAR images powered by large language model (101) interface allow the user to customize the system generated color image by giving the input to the LLM model (105) as claimed in claim 1. This is the novel part of the invention; and
6) Dynamic colorization of SAR images powered by large language model (101) interface system uses the colored image as input for the LLM model to get the textual insights (108) and both image and the text content is stored in the database.

Documents