Roof Segmentation And Light Capacity Prediction Based On Satellite Remote Sensing Images

Roof photovoltaic power generation is rapidly developing due to its low cost and advantageous geographic location.However,in order to ensure the efficiency and stability of roof photovoltaic power generation,accurate prediction and calculation of the available area of the roof and the power generation capacity of the photovoltaic panels are required.In this article,a method based on satellite images and deep learning is improved using remote sensing image analysis.The WHU building dataset is used and data augmentation techniques,including Gaussian noise,random flipping,and blurring,are applied to expand the training dataset.The improved Deeplabv3+ model with attention mechanism added to the ASPP module is used to further improve the segmentation accuracy of the model.The 3×3dilated convolution in the ASPP module is replaced with 1×3 and 3×1 convolutions to speed up model training and inference.Additionally,upsampling layers are added to the decoder module for feature fusion to improve the resolution of the segmentation results.Experimental results show that the improved method can effectively segment the roofs of buildings,with a segmentation accuracy of 89.98%.The predicted results match the actual situation,indicating that this method can accurately predict the installation area of photovoltaic panels.To predict the power generation capacity using real-time monitoring data for photovoltaic power generation,a dataset containing 10 features,including panel temperature,on-site temperature,light intensity,conversion efficiency,voltage,current,power,wind speed,and wind direction,is used.These features play a crucial role in the process of photovoltaic power generation.To ensure data quality,abnormal values in the dataset are processed using forward value replacement,making the dataset more stable and reliable.In addition,a feature correlation analysis is performed on the processed dataset to understand the correlation between different features and select the most relevant features as inputs to the model to improve its prediction accuracy and robustness.For model selection,a neural network model based on CNN-Bi LSTM is used for predicting the photovoltaic capacity.This model combines convolutional neural networks and bidirectional long short-term memory networks,which can effectively capture the nonlinear features of time series data and adaptively adjust the network’s weights and biases to suit different data inputs and prediction requirements.According to the comparison of the predicted results with the actual detection data,the model has high prediction accuracy and generalization ability for predicting photovoltaic capacity.This research has a certain reference value for the real-time monitoring and prediction of photovoltaic power generation.