I am reaching out to seek your guidance regarding a project I am currently working on. The project involves the recognition and classification of drill cuttings using machine and deep learning techniques. As I navigate through this project, I find myself facing a challenge in selecting an appropriate neural network architecture given the unique characteristics of the data.
For the initial step, I have successfully segmented the drill cuttings in the images using image processing techniques. The next phase involves the design of a neural network that can accurately recognize and classify the drill cuttings into five different classes or more. However, I am facing a challenge due to the fact that the drill cuttings are touching each other, and there is no distinct background.
In order to provide you with a better understanding of the data and the challenges involved, I am attaching a few examples of the segmented drill cuttings along
I don't know if i need to use instance segmentation? and whish model i use ? or sementic segmentation like U-net? enter image description here
I have researched and come across some potential options such as U-Net, Mask R-CNN, and various semantic/instance segmentation models. However, I would greatly appreciate your expert opinion on which approach might be most effective for this particular scenario. If you could kindly provide me with some guidance on the appropriate neural network architecture and any other related considerations, I would be extremely grateful.
