Image segmentation in the scanned wooden logs is essential to further process and generate 3D structure of the wooden-logs where insects have removed the wood. The deep learning framework - u-net proposed to generate segmentation maps resulted in more than expected time and resources to train the model. The results obtained with this approach were less segmentation and more of noisy data along the edges of the image. In future, more images can be annotated, and model can be trained by changing epochs, loss functions and other training parameters or by adding more hidden layers to the model to check what is the best fit for this data. Test results obtained from approach II proposed in this report comparatively gave better segmentation output than approach I. Results obtained for pine wood data had over segmented image consolidating the growth rings and other wood part. This approach can be further improved for pine wood data.