高级检索

    钱文秀, 常青, 向辉, 康文斌. 基于深度监督显著目标检测的草莓图像分割[J]. 华东理工大学学报(自然科学版), 2020, 46(1): 114-120. DOI: 10.14135/j.cnki.1006-3080.20181205004
    引用本文: 钱文秀, 常青, 向辉, 康文斌. 基于深度监督显著目标检测的草莓图像分割[J]. 华东理工大学学报(自然科学版), 2020, 46(1): 114-120. DOI: 10.14135/j.cnki.1006-3080.20181205004
    QIAN Wenxiu, CHANG Qing, XIANG Hui, KANG Wenbin. Strawberry Image Segmentation Based on Deeply Supervised Saliency Detection[J]. Journal of East China University of Science and Technology, 2020, 46(1): 114-120. DOI: 10.14135/j.cnki.1006-3080.20181205004
    Citation: QIAN Wenxiu, CHANG Qing, XIANG Hui, KANG Wenbin. Strawberry Image Segmentation Based on Deeply Supervised Saliency Detection[J]. Journal of East China University of Science and Technology, 2020, 46(1): 114-120. DOI: 10.14135/j.cnki.1006-3080.20181205004

    基于深度监督显著目标检测的草莓图像分割

    Strawberry Image Segmentation Based on Deeply Supervised Saliency Detection

    • 摘要: 草莓图像的分割效果直接影响着草莓采摘机器人的实时作业,而目前的草莓图像分割算法大多只研究完全成熟及无遮挡情况下的草莓分割,无法实现草莓的多级分类及遮挡检测。本文提出了一种结合显著性区域检测的草莓图像分割方法,可适用于不同明暗环境,并且有效地实现了对不同成熟度草莓的检测与分割,为后续实现多级分类提供了良好的数据支持。首先,使用限制对比度自适应直方图均衡化处理方法克服了实际采摘图像光线昏暗导致的分割困难;然后,使用结合短连接的整体嵌套显著目标检测算法,利用浅层侧面输出包含丰富细节优势及深层侧面输出定位显著区域优势,从而产生密集且准确的显著区;最后,将显著性区域作为Grabcut的前景进行分割,从而实现草莓图像的准确分割。实验数据及结果表明,本文方法在实际采摘中的遮挡及不同明暗环境下均可获得稳定而准确的分割结果。

       

      Abstract: The development of strawberry picking robot and the information of strawberry growth status play important roles in the process of automated strawberry planting. The segmentation of strawberry image directly affects the strawberry picking robot real-time operation and the analysis on strawberry growth status. All of these require that the strawberry segmentation algorithms should be suitable for different light and dark environments and attain better segmentation effect on the strawberry in different maturity. However, most of the existing algorithms on strawberry image segmentation only considered the segmentation in fixed background or fully mature such that they could not achieve the multi-level classification and the detection under occlusion. This paper proposes a strawberry image segmentation method by combining salient region detection. The contrast-limited adaptive histogram equalization processing method is firstly adopted to increase the contrast between the strawberry and the background under different light intensities. It is well known that the shallow side output contains rich details and deep side output can attain accurate regional position. By means of these above features, this paper applies the Holistically-Nested salient object detection method with short connections to produce the dense and accurate salient areas. And then, these salient areas are segmented as the foreground of Grabcut so as to realize the accurate segmentation of strawberry image. Finally, it is shown from the experiment results that the proposed algorithm can obtain stable and accurate segmentation results under the occlusion and different light and dark environments.

       

    /

    返回文章
    返回