Automatic image analysis applied to the recognition of quartz surface microtextures using neural network

Area of Science:

• Geology
• Sedimentology
• Geomorphology

Background:

• Quartz grain surface microtextures offer insights into sedimentary processes and environmental conditions.
• Microtextural analysis is crucial for sediment provenance studies.
• Subjectivity in manual microtextural recognition can limit accuracy.

Purpose of the Study:

• To develop an automated software solution for identifying quartz grain microtextures.
• To minimize subjectivity and enhance the efficiency of microtextural analysis.
• To introduce DeepGrain, a novel software tool for sediment provenance research.

Main Methods:

• Utilized the DeepLabV3 model for image analysis.
• Applied advanced techniques to improve the DeepLabV3 model's performance.
• Developed the DeepGrain software for automatic microtexture identification.

Main Results:

• Achieved 99% accuracy in identifying the area of tested quartz grains.
• Reached 63% accuracy in identifying mechanical features on grain surfaces.
• Reduced the average analysis time per SEM image to 3.10 seconds.

Conclusions:

• DeepGrain offers a highly accurate and efficient automated method for quartz microtexture analysis.
• The software significantly streamlines the process of sediment provenance studies.
• DeepGrain provides a valuable tool for researchers in sedimentology and geology.