College of Mechanical and Electrical Engineering,Northeast Forestry University,Harbin 150040,China
Show less
文章历史+
Received
Published
2025-03-24
2025-11-15
Issue Date
2025-12-24
PDF (3192K)
摘要
针对落叶松活立木心材精准识别难题,提出一种基于电阻断层成像(electrical resistivity tomography,ERT)的智能分割方法。通过对比分析落叶松样本ERT图像与物理切面,发现心边材的交界区域电阻变化率达90%~94%。根据该阈值范围,建立落叶松活立木心材ERT图像分割标注准则。由于ERT图像获取困难,构建Mini-200小样本和Mids-3200大样本2个数据集,通过小训练集快速适应分割任务,结合大训练集提升模型鲁棒性、减少过拟合。提出改进的语义分割网络(DeepLabv3+)模型,通过引入ResNet101、混合注意力机制(convolutional block attention module,CBAM)与特征强化模块(data communication module,DCM)优化其特征提取能力。消融试验表明,在Mini-200数据集上,改进模型准确率(accuracy,A)、精确率(precision,P)、交并比(intersection over union,IoU)、平均交并比(mean intersection over union,mIoU)和Dice损失函数5个评价指标较基础模型提升0.14%~0.44%;在Mids-3200数据集上,改进后的DeepLabv3+模型分割性能最优,与原模型相比,心材的像素准确率(pixel accuracy,PA)和IoU分别提高了0.32%和2.45%,类别像素准确率(class pixel accuray,CPA)、mIoU和Dice系数分别提高0.47%、2.13%、0.25%,IoU达98.80%,证明改进模型用于落叶松ERT图像心材分割的效果良好。
Abstract
Aiming at the difficulty of accurately recognizing the heartwood of larch standing trees, this study proposes an intelligent segmentation method based on electrical resistance tomography (ERT). By comparing and analyzing ERT images and physical sections of larch samples, it is found that the resistance change rate of the junction region of the heart sapwood reached 90%-94%. Based on this threshold range, the segmentation annotation criterion for ERT image segmentation of larch standing tree heartwood was established. Due to the difficulty in acquiring ERT images, two datasets, Mini-200 small samples and Mids-3200 large samples, were constructed to quickly adapt to the segmentation task through the small training set, and combined with the large training set to improve the model robustness and reduce overfitting. The improved semantic segmentation network (DeepLabv3+) model was proposed to optimize its feature extraction capability by introducing ResNet101, convolutional block attention mechanism (CBAM) and data communication module (DCM). The ablation experiments showed that the five evaluation indexes of accuracy (A), precision (P), intersection over union (IoU), mean intersection over union (mIoU) and Dice loss function of the improved model were improved by 0.14%-0.44% compared with the base model on the Mini-200 dataset; on the Mids-3200 dataset, the improved DeepLabv3+ model had the optimal segmentation performance, and compared with the original model, the pixel accuracy (PA) and IoU of the heartwood were improved by 0.32% and 2.45%, respectively, and the class pixel accuracy (CPA), mIoU and Dice coefficient were improved by 0.47%, 2.13%, 0.25%, respectively, and the IoU reached 98.80%, compared with the original model. It proves that the improved model works well for the segmentation of the heartwood of the ERT image of larch.
采用准确率(accuracy,A)、精确率(precision,P)、交并比(intersection over union,IoU,式中为IoU)、平均交并比(mean intersection over union,mIoU,式中为mIoU)和Dice损失函数5个指标对模型进行评价。前4个指标计算公式为
YANGL, FANGC R, ZHANGJ,et al.The identification of Dalbergia odorifera and D.tonkinensi by gas chromatography mass spectrometry[J].Journal of Nanjing Forestry University (Natural Sciences Edition),2016,40(1):97-103.
[3]
HACHLAFI NEL, LAKHDARF, KHOUCHLAAA,et al.Health benefits and pharmacological properties of hinokitiol[J].Processes,2021,9(9):1680.
[4]
DIVAKARAB N, CHAITHRAS.Electric resistance tomograph (ERT):A review as non-destructive tool (NDT) in deciphering interiors of standing trees[J].Sensing and Imaging,2022,23(1):18.
YANGL L, DONGX B, XUH D.Comparison between electrical resistance method and stress wave method in the detection of decay defects in living trees[J].Forest Engineering,2022,38(4):82-88.
[7]
GUYOTA, OSTERGAARDK T, LENKOPANEM,et al.Using electrical resistivity tomography to differentiate sapwood from heartwood:Application to conifers[J].Tree Physiology,2013,33(2):187-194.
[8]
URSELLAE.In-line industrial computed tomography applications and developments[D].Venezia,Italy:Università Ca'Foscari Venezia,2021.
[9]
HATTAH R, NURDIATIS, HERMADII,et al.Grade classification of agarwood sapwood using deep learning[J].International Journal on Informatics Visualization,2024,8(4):2075-2081.
ZHANGL F, TONGT, XIAOK.Flow pattern identification of gas-liquid two-phase flow based on deep forest and electrical resistance tomography[J].Acta Metrologica Sinica,2023,44(6):893-898.
[12]
VINCIGUERRAA, ALEARDIM, MADSENL M,et al.Stochastic inversion of time-lapse electrical resistivity tomography data by means of an adaptive ensemble-based approach[J].Geophysical Prospecting,2024,72(1):268-284.
SHIX L, WANGL H, XUH D,et al.Quantitative characterization of decay detection in standing trees of Korean pine based on Resistograph and ERT[J].Journal of Beijing Forestry University,2017,39(9):102-111.
[19]
BIEKERD, RUSTS.Non-destructive estimation of sapwood and heartwood width in Scots pine (Pinus sylvestris L.)[J].Silva Fennica,2010,44(2):267-273.
[20]
GUYOTA, OSTERGAARDK T, LENKOPANEM,et al.Using electrical resistivity tomography to differentiate sapwood from heartwood:Application to conifers[J].Tree Physiology,2013,33(2):187-194.
[21]
CZAJKAM, FABISIAKB, FABISIAKE.Emission of volatile organic compounds from heartwood and sapwood of selected coniferous species[J].Forests,2020,11(1):92.
[22]
CHOLLETF.Xception:Deep learning with depthwise separable convolutions[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).July 21-26,2017,Honolulu,HI,USA.IEEE,2017:1800-1807.
[23]
SIY Z, XUH Y, ZHUX Z,et al.SCSA:Exploring the synergistic effects between spatial and channel attention[J].Neurocomputing,2025,634:129866.
[24]
YEUNGM, SALAE, SCHÖNLIEBC B,et al.Focus U-Net:A novel dual attention-gated CNN for polyp segmentation during colonoscopy[J].Computers in Biology and Medicine,2021,137:104815.
[25]
LINT Y, GOYALP, GIRSHICKR,et al.Focal loss for dense object detection[C]//2017 IEEE International Conference on Computer Vision (ICCV).October 22-29,2017,Venice,Italy.IEEE,2017:2999-3007.