Algorithm 1: Adaptive Joint Learning Semantic Segmentation Based on Prior Knowledge

Input: Infrared grayscale image Img; Acquisition time t; Prior threshold

Output: Segmentation results Binary_Img

Step 1: Read the maximum and minimum temperatures corresponding to the pixel intensity extremes of Img.

1: max (Img) -> T_max; min (Img) -> T_min

Step 2: Convert Img to a temperature matrix TI based on T_max and T_min.

2: TI = (Img − min (Img))/(max (Img) − min (Img)) × (T_max − T_min)

Step 3: Calculate dam surface temperature threshold Tthreshold and bias Tbias based on t and the prior threshold.

Step 4: Calculate the proportion of high-temperature anomalous pixels (Pixel_proportion) in TI exceeding Tthreshold + Tbias.

3: for i = 0 to TI.shape [0] do

4:   for j = 0 to TI.shape [1] do

5:      if TI [i, j] > Tthreshold+ Tbias do sumTi = sumTi + 1

6: Pixel_proportion = sumTi/(TI.shape [0] × TI.shape [1])

Step 5: Determine anomaly type.

7: if Pixel_proportion > 99%: global anomaly exists, Binary_Img[:,:] = 255

8:    else if Pixel_proportion < 1%: anomaly-free, Binary_Img[:,:] = 0

9:      else local anomaly exists, do Step 6

Step 6: Remove low-temperature pixel interference.

10: if (Tthreshold − 2 × Tbias) > T_min: TI [TI < (Tthreshold − 2 × Tbias)] = Tthreshold − 2 × Tbias

11: Img = (TI − T_min)/(T_max − T_min) × (max (Img) − min (Img))

Step 7: Perform Otsu’s thresholding for local hollowing segmentation.

Gthreshold, Binary_Img = OTSU(Img)

Step 8: Assign semantic labels to high-temperature anomaly pixels.

Step 9: Apply morphological operations to refine segmentation.

12: Binary_Img = morphologyEx (Binary_Img, cv2.MORPH_OPEN, (3 × 3))

13: Binary_Img = morphologyEx (Binary_Img, cv2.MORPH_CLOSE, (3 × 3))

End