Novel Mask R-CNN Based Mushroom Cluster Tracking in Time-Lapse Images From a Farm Environment

This study aims to develop a novel deep learningbased approach to support the automated mushroom growth monitoring using an object tracking algorithm in conjunction with instance segmentation models. For this purpose, a comprehensive time-lapse image dataset was captured at a farm growing oyster mushrooms. Three different CNN-based and Visual Transformer-based backbone feature extraction components for Mask R-CNN were tested to identify clusters of oyster mushrooms in the images. The clusters recognized by the instance segmentation model, were linked across a series of timelapse images to track their growth across the consecutive images. The tracking function was performed by an 'intersection over union' based object tracking algorithm. This method considers the orientation, size, shape and growth phase of the clusters and accounts for the variance in object detection inherent to the instance segmentation model. The pixelwise size of the mushroom clusters was then used to approximate their growth. The deep instance segmentationbased tracking algorithm adequately identified mushroom clusters of varied size along their production cycle.