Development of activity recognition models for mechanical fuel treatments using consumer-grade GNSS-RF devices and lidar

Abstract

Abstract Mobile technologies are rapidly advancing the field of forest operations and providing opportunities to quantify management tasks in new ways through increased digitalization. For instance, devices equipped with global navigation satellite system and radio frequency transmission (GNSS-RF) enable real-time data collection and sharing of positional data in remote, off-the-grid environments where cellular and internet availability are otherwise inaccessible. In this study, consumer-grade GNSS-RF data were evaluated to determine their effectiveness in developing activity recognition models for excavator-based mastication operations. The ability to automate the classification of cycle elements for operations is valuable for quickly and efficiently quantifying production rates for research and industry applications. The GNSS-RF-based activity recognition model developed successfully classified productive elements versus delay with over 95 per cent accuracy. Individual cycle elements were classified with an overall model accuracy of 73.6 per cent, with individual element classification accuracy ranging from 51.3 per cent for walk/reposition to 95.6 per cent for mastication elements. Reineke’s stand density index, basal area (m2 ha−1) of treated areas and the duration of cycle elements impacted the classification accuracy of the activity recognition model. Impacts of forest stand characteristics on the production rate of mastication treatments were also assessed. Production rates (ha·hr−1) for mastication treatments were affected by the basal area of treated areas. However, the degree to which this would impact operations in practice is minimal. Determining the proper application and capabilities of mobile technologies and remote sensing for quantifying forest operations is valuable in continuing the innovation and advancement of forest digitalization.