Key points are not available for this paper at this time.
In forest operations, established time-study methods, such as the use of a stopwatch and video recording, have dominated for several years. Advancements in machine learning and innovative data loggers present opportunities to reconsider and enhance these methods. This study utilized a neural network algorithm to learn from and generalize multimodal data collected by noise dosimeters and accelerometers placed at two different locations. It involved the implementation of seven distinct neural network models aimed at recognizing operator activities during chainsaw work. The classification accuracy (CA) of distinguishing between “work” and “other” work events reached impressive levels with various combinations of datasets. Notably, the poorest performance was associated with data from the operator’s back-mounted accelerometer alone (CA = 63%), whereas the best results emerged from combining data from the accelerometer on the chainsaw with a noise dosimeter (CA = 99.8%). This research highlights the viability of using accelerometers and noise data to analyze chainsaw operations, suggesting that defined network architectures and parameters can efficiently handle large datasets. This capability will facilitate the documentation of efficiency and delays in cross-cutting and delimbing activities without requiring specialized knowledge of forest operations or time studies. Data collection can span several working days and cover multiple locations, thereby minimizing the time spent on data management and analysis. Future developments in this model could enable a more granular analysis of the chainsaw work duration, further improving our understanding of operational efficiency in forestry.
Borz et al. (Wed,) studied this question.