Due to the defects such as low resolution, lack of hierarchy, and blurred visual effects in infrared images, the accuracy of detecting infrared images is much lower than that of detecting visible images. In this study, we propose a super-resolution enhancement method for single-frame infrared images based on improved Real-ESRGAN to resolve the problem of low resolution and lack of detailed texture of infrared images so that we can improve the accuracy of object detection. We add attention mechanism based on the original network to improve the network’s ability to extract the detailed texture. In addition, we also adjust the training epochs of the generator and discriminator to accelerate the generator update and prevent the discriminator from converging too fast. We also use pooling layers for downsampling to remain the important detailed texture features and make it easier for convolution layers to extract detailed features. The experimental results show that compared with original Real-ESRGAN, when improving the resolution twice as before, our improved network reach an increment of 6.23% of PSNR and 13.9% of SSIM, and when improving the resolution four times as before, the increment is 0.95% of PSNR and 4.42% of SSIM. In object detection, we use YOLOv5 to detect super-resolution infrared images generated by our improved network and the original infrared images and reach an increment of 2.95% of mAP. These promising results confirm that our network works effectively as a method of infrared image super-resolution enhancement and improves object detection accuracy.
Mugwort floss, valued in traditional Chinese medicine, varies in therapeutic properties and market price based on origin and production year. Traditional identification methods, due to their destructiveness and low accuracy, often confuse mugwort floss with A.stolonifera and cause a testing waste. Hyperspectral Imaging, a non-contact technique, offers potential for rapid identification of such medicinal materials. In this paper, we explore hyperspectral data to differentiate mugwort and A.stolonifera using deep learning and neural networks. Using a massive hyperspectral dataset from mugwort and wormwood from two regions across four years, we analyzed performance using metrics like Accuracy, Specificity, and F1 Score. The self-attention-based Backpropagation Neural Network model showed the most promising results for accurate classification. This approach has potential future applications in various fields using Hyperspectral data
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.