Quaternary Classification of Image Fog Opacity on Arduino Nano BLE using Modified CNN

Image dehazing is a significant area of study with many real-world applications. However, only foggy/hazy images can be used with this technology. Therefore, to use image dehazing technology, it is necessary to classify images as either hazy/foggy or clear to eliminate computational redundancy. It is not optimal to rely solely on human perception because it is unpredictable and vulnerable to bias. Therefore, an appropriate algorithm is necessary for effective image classification. An initial haze-clear classification is re- /sub-classified into more classes. The algorithm presented in this research study classifies images into four categories: clear, dense fog, hazy, or light fog. Tiny Machine Learning Shield and Arduino Nano 33 BLE are used to implement the algorithm. With the aid of Edge Impulse, the Convolutional Neural Network (CNN) model is trained, examined, and approved. Real-time image capture by the camera is combined with classification by the model. On the serial monitor, the classification results are shown.