Image-based air quality estimation using convolutional neural network optimized by genetic algorithms: A multi-dataset approach

Air pollution poses significant threats to human health and the environment, making effective monitoring increasingly essential. Traditional methods using fixed monitoring stations have challenges related to high costs and limited coverage. This paper proposes a new approach using convolutional neural networks with genetic algorithms for estimating air quality directly from images. The convolutional neural network is optimized using genetic algorithms, which dynamically tune hyperparameters such as learning rate, batch size, and momentum to improve performance and generalizability across diverse environmental conditions. Our approach improves performance and reduces the risk of overfitting, thus ensuring balanced and robust results. To mitigate overfitting, we implemented dropout layers, batch normalization, and early stopping, significantly enhancing the model’s generalization capability. Specifically, three different open-access datasets were combined into a single training dataset, capturing extensive temporal, spatial, and environmental variability. Extensive testing of the model performance was conducted with a broad set of metrics, including precision, recall, and F1 score. The results demonstrate that our model not only achieves high accuracy but also maintains well-balanced performance across all metrics, ensuring robust classification of different air quality levels. For instance, the model achieved a precision of 0.97, a recall of 0.97, and an overall accuracy of 0.9544 percent, outperforming baseline methods significantly in all metrics. These improvements underscore the effectiveness of Genetic Algorithms in optimizing the model.