A Novel Method for Shadow Detection in PV Module Using Deep Learning Approach

Photovoltaic (PV) systems are often affected by dust, dirt, and obstructions such as trees and buildings, leading to partial shading and uneven sunlight distribution. This results in significant energy losses and impacts the electrical performance of PV generators. Accurate shadow boundary extraction is crucial for inspecting and monitoring PV panels, as it helps define module boundaries clearly for developers and maintenance providers. This article introduces a novel method for extracting shadow boundaries in PV plants using a deep neural network designed for shadow detection. Leveraging a shading dataset, the approach incorporates a spatial attention module to enhance spatial info extraction from feature maps. This module effectively captures semantic details both horizontally and vertically, addressing challenges in identifying small or ambiguous shadow areas and minimizing missed or incorrectly detected shadows. The results validate that the trained model distinguish the borders of shaded PV module with an accuracy of 99.33 %.