Local feature descriptor for multispectral image matching of a large-scale pv array

• Main Authors: Tan, Li Ven, Mohd Shawal, Jadin, Kamarul Hawari, Ghazali, Ahmad Syahiman, Mohd Shah, Muhammad Khusairi, Osman
• Format: Conference or Workshop Item
• Language: English; English
• Published: Institute of Electrical and Electronics Engineers Inc. 2022
• Subjects: T Technology (General); TA Engineering (General). Civil engineering (General); TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://umpir.ump.edu.my/id/eprint/39433/1/Local%20Feature%20Descriptor%20for%20Multispectral%20Image%20Matching%20of%20a%20Large.pdf; http://umpir.ump.edu.my/id/eprint/39433/2/Local%20feature%20descriptor%20for%20multispectral%20image%20matching%20of%20a%20large-scale%20pv%20array_ABS.pdf; http://umpir.ump.edu.my/id/eprint/39433/; https://doi.org/10.1109/I2CACIS54679.2022.9815491

Possible faults in the photovoltaic modules must be detected early in order to preserve their long-term reliability while maximizing power output. Aerial thermal image inspection is frequently used to detect and locate photovoltaic module hotspots. However, noises can make it difficult to detect a hotspot from this image, causing the hotspot to be incorrectly located due to thermal reflection from the environment. Examining both visual and thermal images of photovoltaic modules appears to be one of the solutions. The multispectral image matching of photovoltaic modules is presented in this paper. The absolute structure map (SMi) and the directional structure map (DSMi) are proposed. The local region of each interest point is then described using a histogram of the oriented gradient based on the SMi and DSMi. For the SMi, the Gabor wavelet filter is applied, whereas the average filter is applied to the DSMi for the construction of the histogram bins. Finally, the normalized feature vectors are merged. Experiments were carried out to evaluate the performance of the proposed structure map feature descriptor. According to the findings, this approach could give precision and recall up to 0.82 and 0.97 respectively.