無人機(jī)海上艦船目標(biāo)影像超分辨率重建

摘" 要： 針對無人機(jī)在獲取海上艦船目標(biāo)影像時面臨的實(shí)時性與清晰度之間的矛盾，提出一種影像壓縮模糊重建方法。該方法利用改進(jìn)的YOLOv8檢測模型和Real?ESRGAN網(wǎng)絡(luò)，通過數(shù)據(jù)集構(gòu)建、網(wǎng)絡(luò)訓(xùn)練調(diào)試和部署運(yùn)用等步驟，實(shí)現(xiàn)了在有限帶寬和計算資源環(huán)境下地面端高質(zhì)量艦船目標(biāo)影像的實(shí)時重建。首先利用改進(jìn)的YOLOv8模型對影像中艦船目標(biāo)進(jìn)行精準(zhǔn)檢測和定位，隨后通過Real?ESRGAN網(wǎng)絡(luò)對壓縮及模糊影像進(jìn)行重建，以恢復(fù)影像的高分辨率和細(xì)節(jié)信息。實(shí)驗(yàn)結(jié)果表明，該方法不僅顯著提升了影像的清晰度和檢測準(zhǔn)確性，還大幅減少了帶寬消耗，滿足了無人機(jī)艦船識別的高實(shí)時性要求，且在資源受限的情況下表現(xiàn)尤為突出。為無人機(jī)在海上艦船目標(biāo)監(jiān)測領(lǐng)域提供了一種有效的解決方案，不僅提高了無人機(jī)的監(jiān)測和識別能力，也為進(jìn)一步推進(jìn)無人機(jī)在海洋監(jiān)測中的廣泛應(yīng)用奠定了基礎(chǔ)。

關(guān)鍵詞： 無人機(jī)影像； 海面艦船； 雙向特征融合模型； Real?ESRGAN網(wǎng)絡(luò)； 改進(jìn)的YOLOv8檢測模型； 海上艦船目標(biāo)監(jiān)測

中圖分類號： TN911.73?34； TP751" " " " " " " " " "文獻(xiàn)標(biāo)識碼： A" " " " " " " " " 文章編號： 1004?373X（2025）01?0017?06

Super?resolution reconstruction of UAV maritime vessel target images

SUN Weiwei1， CUI Yaqi1， 2， 3， ZHANG Shaoqing2， 3， XIA Shutao1

（1. Naval Aeronautical University， Yantai 264000， China;

2. Shenyang Aircraft Design and Research Institute of Aviation Industry of China， Shenyang 110035， China;

3. Northwestern Polytechnical University， Xi’an 710072， China）

Abstract： A method for compressive and blurry image reconstruction has been proposed to get rid of the conflict between real?time requirements and image clarity during the acquisition of maritime vessel images by unmanned aerial vehicles （UAVs）. By utilizing an improved YOLOv8 detection model and Real?ESRGAN network， this method achieves real?time reconstruction of high?quality vessel images at the ground station under limited bandwidth and computational resource constraints with the steps of dataset construction， network training， debugging and deployment. Initially， the improved YOLOv8 model is used for precise detection and localization of vessel within the images. Subsequently， the Real?ESRGAN network is used to reconstruct the compressive and blurry images to restore high?resolution and details of the image. Experimental results indicate that the method enhances image clarity and detection accuracy significantly while greatly reducing bandwidth consumption， meeting the high real?time requirements of UAV?based vessel recognition， particularly in resource?constrained scenarios. This method provides an effective solution for UAVs in the field of maritime vessel monitoring， enhancing their capabilities for surveillance and identification， and laying the groundwork for the broader application of UAVs in marine monitoring.

Keywords： UAV image; surface vessel; bidirectional feature fusion model; Real?ESRGAN network; improved YOLOv8 detection model; monitoring of maritime vessel target

0" 引" 言

無人駕駛飛機(jī)在民用和軍事應(yīng)用中都承擔(dān)著重要的任務(wù)，為有效支撐海上艦船目標(biāo)發(fā)現(xiàn)識別任務(wù)，要求無人機(jī)盡可能遠(yuǎn)地獲取艦船目標(biāo)高清影像[1]。然而由于通信帶寬的限制，地面站僅能獲取壓縮后的模糊影像[2]，無法獲取高清無損影像，嚴(yán)重影響地面站影像分析判讀工作，海上目標(biāo)檢測識別作為無人機(jī)的重要應(yīng)用方向，實(shí)現(xiàn)無人機(jī)海上艦船目標(biāo)影像壓縮模糊重建需求迫切。……