FPNC Net: A hydrogenation catalyst image recognition algorithm based on deep learning.

Area of Science:

• Chemical Engineering
• Computer Vision
• Materials Science

Background:

• Accurate identification of hydrogenation catalysts is crucial for the chemical industry.
• Existing methods struggle with recognition accuracy due to catalyst adhesion and stacking.
• Efficient screening of high-performance catalyst carriers requires intelligent image recognition solutions.

Purpose of the Study:

• To develop an intelligent image recognition algorithm for hydrogenation catalysts.
• To overcome the limitations of low recognition accuracy caused by catalyst adhesion and stacking.
• To enhance the efficiency of screening high-performance catalyst carriers.

Main Methods:

• Proposed an image recognition algorithm based on FPNC Net.
• Utilized Resnet50 backbone network for feature extraction.
• Employed spatially-separable convolution kernels for multi-scale feature extraction.
• Integrated Feature Pyramid Network (FPN) for deep and shallow feature fusion.
• Incorporated an attention module for adaptive weight adjustment.

Main Results:

• The FPNC Net model achieved a recognition accuracy of 94.2%.
• Demonstrated a significant improvement in Average Precision (AP) by 19.37% compared to the original CenterNet model.
• The enhanced model showed a substantial increase in detection accuracy for hydrogenation catalyst targets.

Conclusions:

• The developed FPNC Net algorithm effectively addresses challenges in hydrogenation catalyst image recognition.
• The model exhibits high capability and significant enhancement in detecting hydrogenation catalyst targets.
• This advancement aids researchers in efficiently screening high-performance catalyst carriers.