To improve the accuracy of component temperature analysis and heating fault detection in infra-red images of power equipment

a thermal state detection method integrating instance segmentation and temperature matrix analysis was proposed. Infrared images and corresponding temperature matrices were used as the data source

and a power equipment component segmentation model based on object detection and instance segmentation was constructed. A global context network ( GCNet) was introduced into the backbone network to enhance contextual feature modeling

while an omni-dimensional dynamic convolu-tion ( ODConv) structure was incorporated into the neck network to reduce feature redundancy and im-prove multi-scale feature fusion. Based on the segmented component regions

the corresponding tempera-ture matrix data were extracted

and statistical analysis of component temperature distribution was con-ducted. Combined with fault temperature criteria of power equipment

preliminary diagnosis of heating faults was achieved. Experimental results show that the improved model achieves mean average precision values of 89. 7% for mean average precision( mAP)

mAP@0. 5( M) and 57. 9% for mAP@0. 5∶0. 95( M)

representing improvements of 2. 6% and 3. 6%

respectively

compared with the baseline model. In tem-perature distribution analysis

the standard deviation of component temperature is reduced by 2. 58

and kurtosis is increased by 3. 4

indicating a higher degree of temperature data concentration. These results demonstrate that the proposed method effectively improves fault detection and temperature analysis accura-cy in infrared images of power equipment

providing reliable support for equipment condition monitoring and fault diagnosis.