https://github.com/gokhuntokay/Code-Data-Materials

In this study, four AI models “Grok 3, ChatGPT 4.0/Turbo, Deepseek R1, and Gemini 2.5 Pro” were given identical prompts to generate Python scripts for analyzing 25 grayscale SEM images of noodle samples. The analysis focused on: Connected Components (Threshold=50, THRESH_BINARY_INV), Contour Detection (Threshold=50, THRESH_BINARY_INV, cv2.RETR_EXTERNAL), Black pixels=Pore% (Threshold=10, THRESH_BINARY), threshold values were determined by preliminary tests in Imagej. Each script was run in an online Python environment (e.g., Google Colab), where the images were stored in the “/content/”directory. Required libraries were installed before execution (Opencv-Python, numpy, pandas, etc.). The following metrics were calculated: Number of Particles (NOP): Total number of detected particles, Total Area (TA): Combined area of all particles (px²), Average Size (AS): Mean particle area (TA/NOP, px²), Area % (A): Percentage of image area covered by particles, Pore % (P): Percentage of black pixels indicating pores. Results were transferred to Excel for comparison. The Java-based ImageJ-Fiji software was also used for reference measurements. Since pore percentage results were identical in both Python and Java, they were reported only once. Note: Due to imaging characteristics, particles appear darker than the background, so cv2.THRESH_BINARY_INV was used to highlight particles as white. For contour analysis, cv2.RETR_EXTERNAL was chosen to detect only outer boundaries, simplifying the process.