Identifying non-thrive trees and predicting wood density fromresistograph using temporal convolution network

다국어 초록 (Multilingual Abstract)

Deep learning approaches have been adopted in Forestry research including tree classifica-tion and inventory prediction. In this study, we proposed an application of a deep learningapproach, Temporal Convolution Network, on sequences of radial resistograph profiles toidentify non-thrive trees and to predict wood density. Non-destructive resistance drillingmeasurements on South and West orientations of 274 trees in a 41-year-old Douglas-firstand in Marion County, Oregon, USA were used as input series. Non-thrive trees weredefined based on their changes in social status since establishment. Wood density wasderived by X-ray densitometry from cores obtained by increment borers. Data was split forcross validation. Optimal models were fine-tuned with training and validation datasets, thenrun with test datasets for model evaluation metrics. Results confirmed that the applicationof the Temporal Convolution Network on resistograph profiles enables non-thrive tree iden-tification with the probability, represented by the area under the Receiver OperatorCharacteristic curve, equal to 0.823. Temporal Convolution Network for wood density predic-tion showed a slight improvement in accuracy (RMSE¼18.22) compared to the traditionallinear (RMSE¼20.15) and non-linear (RMSE¼20.33) regression methods. We suggest thatthe use of machine learning algorithms can be a promising methodology for the analysis ofsequential data from non-destructive devices.

참고문헌 (Reference)

1 Todoroki CL, "Wood density estimates of standing trees by micro-drilling and other nondestructive measures" 51 : 6-, 2021

2 Pretzsch H, "The social drift of trees. Consequence for growth trend detection, stand dynamics, and silviculture" 140 (140): 703-719, 2021

3 Abadi M, "TensorFlow:A system for large-scale machine learning" 265-283, 2016

4 Pelletier C, "Temporal convolutional neural network for the classification of satellite image time series" 11 (11): 523-, 2019

5 Yan J, "Temporal convolutional networks for the advance prediction of ENSO" 10 (10): 1-15, 2020

6 Rinn F, "Resistograph and Xray density charts of wood"

7 Iliadis L, "Predicting Douglas-fir wood density by artificial neural networks(ANN)based on progeny testing information" 67 (67): 771-777, 2013

8 Demertzis K, "Machine learning use in predicting interior spruce wood density utilizing progeny test information" 28 (28): 505-519, 2017

9 Chollet F, "Keras"

10 El-Kassaby YA, "In situ wood quality assessment in Douglas-fir" 7 (7): 553-561, 2011

11 Park CY, "Evaluation of specific gravity in post member by drilling resistance test" 34 (34): 1-9, 2006

12 Schmidhuber J, "Deep learning in neural networks : an overview" 61 : 85-117, 2015

13 Wang X, "Decay detection in red oak trees using a combination of visual inspection, acoustic testing, and resistance microdrilling" 34 (34): 1-4, 2008

14 Bouffier L, "Can wood density be efficiently selected at early stage in maritime pine (Pinus pinaster Ait.)?" 65 (65): 106-106, 2008

15 Chantre G, "Can drill resistance profiles (Resistograph) lead to within-profile and within-ring density parameters in Douglas fir wood" 18-22, 1997

16 Weiner J, "Asymmetric competition in plant populations" 5 (5): 360-364, 1990

17 da Silva Oliveira JT, "Assessing specific gravity of young Eucalyptus plantation trees using a resistance drilling technique" 71 (71): 137-145, 2017

18 Hosmer DW Jr., "Applied logistic regression (Vol. 398)" John Wiley & Sons 2013

19 Zeide B, "Analysis of growth equations" 39 (39): 594-616, 1993

20 Bai S, "An empirical evaluation of generic convolutional and recurrent networks for sequence modeling"

21 Curtis RO, "A simple index of stand density for Douglasfir" 28 (28): 92-94, 1982

22 Gao S, "A critical analysis of methods for rapid and nondestructive determination of wood density in standing trees" 74 (74): 27-, 2017