Phenotyping methodologies of log end splitting in eucalyptus (Eucalyptus spp.)

This study addresses the crucial consideration of log end splitting in breeding programmes for treated wood. There is a paucity of research focused on efficiently optimizing the phenotyping process for this particular trait. The study aimed to compare methodologies for log end splitting phenotyping and develop an image-based crack evaluation approach. Initially, 32 eucalyptus clones underwent phenotyping using manual measurement, digital image analysis and visual evaluation. Results showed similar phenotypic values, but image analysis demonstrated better clone discrimination, reducing evaluation time to 78 h compared to manual measurement. The second part focused on testing convolutional neural network architectures (UNet, LinkNet and FPN) using real and synthetic images. U-Net exhibited slight superiority based on higher Intersection over Union (IoU) values, exhibiting a high correlation (.89) with true values. This approach significantly reduced evaluation time to approximately 10.15 h, emphasizing its efficiency compared to traditional methods.