Improving a stem-damage monitoring system for a single-grip harvester using a logistic regression model in image processing

Abstract

Monitoring systems were applied to a single-grip harvester logging cut-to-length roundwood in Finland. Using single-grip harvesters may results in stem damages to the remaining trees during thinning, thereby reducing the growth and wood quality of the trees. These concerns justify the need for a decision support system to monitor stem damage in sustainable wood supply. One method to carry out harvesting-quality monitoring involves the application of image processing. The development of a monitoring system relies on the simulation of stem damage to 54 trees, 23 of which were Scots pine (Pinus sylvestris L.) and 31 of which were Norway spruce (Picea abies Karst). The algorithm was validated using data from 15 stands (463 trees) in the field. The damage to the stem was systematically photographed from a strip road and was intended to simulate the operation of machine vision. To determine the relationship between successful detection and stand-harvesting condition, an analysis of the detection of stem damage was conducted using the image processing technique. Meaningful relationships, which are suitable for use in linear classifiers for image processing, were discovered using logistic regression analysis. To improve the stem-damage monitoring system for a single-grip harvester, it was concluded that given the requirement for accurate thresholds of the stem-damage texture, development should focus on multi-view photogrammetry of the damage using machine learning. The monitoring system could be applicable outside Finland for the quality management of sustainable wood procurement.