亚洲一区欧美在线,日韩欧美视频免费观看,色戒的三场床戏分别是在几段,欧美日韩国产在线人成

基于EnlightenGAN圖像增強(qiáng)的自然場(chǎng)景下蘋果檢測(cè)方法
作者:
作者單位:

作者簡(jiǎn)介:

通訊作者:

中圖分類號(hào):

基金項(xiàng)目:

國(guó)家重點(diǎn)研發(fā)計(jì)劃項(xiàng)目(2019YFD1002401)


Application of Image Enhancement Technology Based on EnlightenGAN in Apple Detection in Natural Scenes
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 圖/表
  • |
  • 訪問(wèn)統(tǒng)計(jì)
  • |
  • 參考文獻(xiàn)
  • |
  • 相似文獻(xiàn)
  • |
  • 引證文獻(xiàn)
  • |
  • 資源附件
  • |
  • 文章評(píng)論
    摘要:

    自然光照下陰影會(huì)降低采摘機(jī)器人視覺系統(tǒng)對(duì)蘋果目標(biāo)的準(zhǔn)確感知能力,導(dǎo)致采摘效率低。本研究采用EnlightenGAN算法進(jìn)行圖像增強(qiáng),以實(shí)現(xiàn)陰影的去除和蘋果目標(biāo)檢測(cè)精度的提升。首先通過(guò)圖像光照歸一化處理得到自正則化注意力圖,達(dá)到圖像陰影檢測(cè)的目的,再采用注意力引導(dǎo)的U-Net作為生成器骨干網(wǎng)絡(luò)得到增強(qiáng)后的圖像,然后通過(guò)全局-局部判別器來(lái)比對(duì)圖像信息,最終在生成器和判別器的對(duì)抗中達(dá)到圖像質(zhì)量增強(qiáng)的效果。為了進(jìn)一步檢驗(yàn)該方法的陰影去除效果,分別采用EnlightenGAN、Zero_DCE、Adaptive_GAMMA、RUAS等算法在MinneApple公共數(shù)據(jù)集上進(jìn)行試驗(yàn)驗(yàn)證。結(jié)果表明,EnlightenGAN算法均方誤差較Zero_DCE、Adaptive_GAMMA、RUAS算法分別降低19.21%、59.47%、67.42%,峰值信噪比增加6.26%、34.55%、47.27%,結(jié)構(gòu)相似度提高2.99%、23.21%、68.29%。同時(shí),在對(duì)果園拍攝的蘋果圖像進(jìn)行標(biāo)注后,將其送入YOLO v5m目標(biāo)檢測(cè)網(wǎng)絡(luò)進(jìn)行蘋果檢測(cè)訓(xùn)練。并對(duì)EnlightenGAN算法增強(qiáng)前后的蘋果圖像進(jìn)行了測(cè)試,圖像增強(qiáng)前后檢測(cè)精確率分別為97.38%、98.37%,召回率分別為74.74%、91.37%,F(xiàn)1值分別為84%、94%,精確率、召回率和F1值分別提升1.02%、22.25%、11.90%。為證明模型有效性,對(duì)不同數(shù)據(jù)集進(jìn)行了試驗(yàn),結(jié)果表明EnlightenGAN算法增強(qiáng)后的目標(biāo)檢測(cè)精確率、召回率和F1值較無(wú)增強(qiáng)算法及Zero_DCE、Adaptive_GAMMA、RUAS算法有顯著提升。由此可知,將EnlightenGAN算法應(yīng)用于蘋果采摘機(jī)器人的視覺系統(tǒng),可以有效克服果園圖像光照不均以及存在陰影的影響,提升果實(shí)目標(biāo)檢測(cè)性能。該研究可為自然條件下復(fù)雜光照環(huán)境中的果實(shí)檢測(cè)提供借鑒。

    Abstract:

    Under natural light conditions, the presence of shadows reduced the accurate perception ability of apple harvesting robot towards apple targets, leading to low picking efficiency. Therefore, an EnlightenGAN algorithm for image enhancement was proposed, which effectively improved the accuracy of shadow removal and apple object detection. This algorithm first obtained a self-regularized attention map through image lighting standardization to achieve image shadow detection. Next, an attention-guided U-Net was used as the backbone network of the generator to obtain the enhanced image. Then, the information before and after enhancement was compared using a global-local discriminator, and image enhancement was ultimately achieved in the confrontation between the generator and discriminator. To further evaluate the effectiveness of the proposed method, EnlightenGAN, Zero_DCE, Adaptive_GAMMA, and RUAS algorithms were tested on the publicly available MinneApple dataset. Compared with Zero_DCE, Adaptive_GAMMA, and RUAS algorithms, the MSE of EnlightenGAN algorithm was decreased by 19.21%, 59.47%, and 67.42%, respectively, while the PSNR was increased by 6.26%, 34.55%, and 47.27%, respectively. The SSIM was increased by 2.99%, 23.21%, and 68.29%, respectively. The detection P of EnlightenGAN algorithm before and after enhancement were 97.38% and 98.37%, respectively, with R of 74.74% and 91.37%. The F1 score were 84% and 94%, respectively. The precision, recall, and F1 score were improved by 1.02%, 22.25%, and 11.90%, respectively. In order to verify the effectiveness of the model, different datasets were tested, and the results showed that the target detection precision, recall and F1 score after the enhancement of the EnlightenGAN algorithm were improved compared with the non enhanced algorithm, Zero_DCE, Adaptive_GAMMA and RUAS algorithms. All results indicated that the proposed method can effectively improve the detection precision under uneven lighting conditions and provide reference for the visual system of apple harvesting robot.

    參考文獻(xiàn)
    相似文獻(xiàn)
    引證文獻(xiàn)
引用本文

宋懷波,楊涵茹,蘇曉薇,周昱宏,高昕怡,尚鈺瑩,張姝瑾.基于EnlightenGAN圖像增強(qiáng)的自然場(chǎng)景下蘋果檢測(cè)方法[J].農(nóng)業(yè)機(jī)械學(xué)報(bào),2024,55(8):266-279. SONG Huaibo, YANG Hanru, SU Xiaowei, ZHOU Yuhong, GAO Xinyi, SHANG Yuying, ZHANG Shujin. Application of Image Enhancement Technology Based on EnlightenGAN in Apple Detection in Natural Scenes[J]. Transactions of the Chinese Society for Agricultural Machinery,2024,55(8):266-279.

復(fù)制
分享
文章指標(biāo)
  • 點(diǎn)擊次數(shù):
  • 下載次數(shù):
  • HTML閱讀次數(shù):
  • 引用次數(shù):
歷史
  • 收稿日期:2024-03-24
  • 最后修改日期:
  • 錄用日期:
  • 在線發(fā)布日期: 2024-08-10
  • 出版日期: