30988 research outputs found
Sort by
Identification and segmentation of branch whorls and sawlogs in standing timber using terrestrial laser scanning and deep learning
No full textGaining insight into the wood quality of standing timber could facilitate more precise utilization of wood material, thereby promoting a more sustainable use of forest resources. In this study, we utilized convolutional neural network–based object detectors to segment individual branch whorls and sawlog sections from images derived from terrestrial laser scanning (TLS) point clouds. TLS was employed to capture the point clouds of 479 Norway spruce sample trees (Picea abies (L.) H. Karst.) from 14 stands in southeastern Finland. Subsequently, the trees were harvested and the sawlogs measured with X-ray at an industrial sawmill. The convolutional neural network–based branch whorl detector was trained with 2D images of the stem sections of trees in the TLS point clouds from which the branch whorls were manually annotated. The sawlog section detector was trained with 2D TLS images of whole trees in which sawlogs were automatically annotated, using the sawmill measurements. Comparing the detections of the whorl detector with those of the X-ray measurements yielded a root-mean-squared error of 7.73 (64.41%) for the whorl count. Additionally, we conducted further comparison of the detections against a dataset in which the whorls were manually measured from the TLS images, resulting in a root-mean-squared error of 3.99 (20.60%). The detections made by the sawlog detector in the TLS images of whole trees were utilized to calculate the predicted log length and volume, which were then compared with the sawmill measurements of the reference logs. In this comparison, the root-mean-squared error of log length was 0.73 m (15.18%), and that of volume was 0.10 m3 (36.62%). The results indicate that the whorl detector can be utilized for extracting branching features of standing timber that can serve as predictors of the internal wood quality. However, directly depicting the internal knot structure with external branch whorl detections poses a challenge. Additionally, while the sawlog detector demonstrated moderate performance in sawlog segmentation, the accuracies of the predicted log length and volume were relatively weak. Nevertheless, we anticipate that deep learning–based approaches can enhance the autonomous characterization of standing timber, e.g. when laser scanners in harvesters become more commonplace
Metsäsuhteet toisinkerrottuna : Moniaistinen ja monilajinen tietäminen uudistuvassa museoympäristössä
No full textSocial and healthcare reform and New Public Management - Analysis of government programs from 2003–2019
No full text
The association between job demands, job resources, and perceived strain among Finnish psychologists
No full text
Reference black-body radiation source for emissivity measurements on the frequency range 3-30 THz
No full textThe emergence of radiation sources in different frequency ranges fuels the demand for techniques for spectrally resolved calibrations of their intensity. Thermal radiation of a perfectly absorbing object (black-body radiation) is used as an etalon in the near-infrared–visible ranges. However, to date, no suitable material exists as a reference for thermal radiation in the frequency range 3–30 THz. In this work, we demonstrate a thermal source that can serve a broad THz frequency range. We employ a moth-eye-structured silicon coated with conductive graphitic film exhibiting absorbance above 99.5%. By using the thermal emission of a silicon-based Salisbury screen heated at a temperature of 60 °C we demonstrate that the moth-eye structure can be used for calibrating THz devices. We also show that the developed approach may allow one to engineer the spectrum of the frequency comb source using a two-layer heterostructure
