Digital image processing methods in industry environment

Abstract

This research is focused on object detection using artificial intelligence in the woodworking industry. In addition, the goal is to compare the performance of object detection where the scene is captured by two different devices – a standard mobile phone camera and the specialized multispectral camera Monarch II. from Unispectral company. We explore the impact of different wavelengths on object detection capabilities, given that the multispectral camera captures scenes outside the visible part of the electromagnetic spectrum. The convolutional neural networks, specifically the YOLO architecture, are used to evaluate our experiments. Challenges arising from differences in resolution and alignment between the two capture devices are addressed through preprocessing techniques. Subsequently, we perform a comparative analysis of various spectral bands applied to two distinct datasets. The first dataset focuses on detecting wooden log cuts, while the second dataset involves the detection of tree trunks within irregular forest terrain. Additionally, we implement a late fusion technique to combine results from multiple spectral bands to enhance detection performance. The goal is to demonstrate the potential of multispectral cameras in the field of computer vision applied to natural materials like wood.