Optimized Smoking Detection using Fine-Tuned EfficientNetB3 in Deep Learning

This research study provides a complete deep learning-based smoking detection method by integrating a fine-tuned EfficientNetB3 model. Public health monitoring depends much on smoking detection, which also provides a possible way to lower smoking-related diseases by early discovery and intervention. This research work aims to use visual image data to precisely characterise smoking and non-smoking behaviours. Drawn from the open-source Kaggle platform, the 320 photos in the dataset include 196 labelled as smoking and 126 as not smoking. Image resizing, normalising, and data augmentation among other preprocessing methods were used to lower overfitting and increase generalisation, thereby improving model performance. Additional layers including batch normalisation, dropout, and dense layers were included to maximise model accuracy and efficiency after much fine-tuning of the EfficientNetB3 design. The model exhibits significant capacity in differentiating smoking from non-smoking circumstances with a general accuracy of 90%. The results reveal that deep learning models such fine-tuned EfficientNetB3 can be helpful tools for smoking detection, so providing tremendous value in health monitoring systems.