Research on Morphological Detection of FR I and FR II Radio Galaxies Based on Improved YOLOv5

Recently, astronomy has witnessed great advancements in detectors and telescopes. Imaging data collected by these instruments are organized into very large datasets that form data-oriented astronomy. The imaging contain many radio galaxies (RGs) interesting to astronomers. However, considering the scale of astronomical databases information age is extremely large, a manual search impractical given need for labor. Therefore, ability detect specific types largely depends on computer algorithms. Applying machine learning algorithms sets can more effectively using photometric images. Astronomers motivated develop tools automatically analyze massive data, including developing an automatic morphological detection specified sources. Galaxy Zoo projects have generated interest visually classifying galaxy samples CNNs. Banfield studied morphologies host derived from visual inspection Radio project. there relatively studies classification, while fewer detection. We model, which realizes location classification Fanaroff–Riley class I (FR I) II II) galaxies. field target also developed rapidly since convolutional neural network was proposed. You Only Look Once: Unified, Real-Time Object Detection (YOLO) neural-network-based model proposed Redmon et al. made several improvements effect dense based original YOLOv5, mainly following. (1) use Varifocal loss, whose function weigh positive negative asymmetrically highlight main sample training phase. (2) Our adds attention mechanism convolution kernel so feature extraction adjust size receptive dynamically deep networks. In this way, our good adaptability identifying different sizes picture. (3) empirical practices suitable small detection, such as image segmentation reducing stride layers. Apart three major contributions novel points thesis included sources, i.e., images optical images, aiming at better performance accurate positioning. used SDSS, FIRST, label FR Is IIs catalogs create set IIs. Subsequently, we train improved YOLOv5 finally realize Experimental results prove method achieves performance. [email protected] reaches 82.3%, (Ra Dec) be identified accurately. significance. For example, it help astronomers find build larger-scale catalog. extended other RGs. Thus, locate type considerably shorter time with minimum human intervention, or combined observation (spectrum redshift) explore properties