Comparitive analysis of CNN algorithms for mushroom classification with proposed lightweight model

The classification of mushroom species presents significant ecologic and health-related challenges; advancement in classification techniques is required to gain reliable identifications. This master's thesis aims to explain a methodology that was devised and evaluated in the development of a novel, lightweight convolutional neural network designed specifically for the task of mushroom classification. The study provides a custom convolutional neural network model computationally cost-effective and capable of high-precision classification, fit for real-time usage. Hence, we evaluate the proposed model on this dataset of curated mushroom images with traditional classifiers and state-of-the-art convolutional neural network architectures, such as EfficientNet-B7, ResNet-50, InceptionV3, and MobileNetV2. The custom model is depth-wise separations engineered in such a way that while they reduce the computational load, they don't compromise the effectiveness of the model. This approach helps the model function effectively even on platforms having low computational resources. A comprehensive evaluation would reveal that a custom convolutional neural network has reasonable accuracy not only in the identification of different mushroom species vis-à-vis existing models but has significantly lightened the classification process as well. This study further strengthens the robustness of the model by integrating advanced regularization techniques to mitigate the problem of overfitting so that the model classifies across different and unseen mushroom species. Finally, it is highlighted in the thesis that the diversity of image backgrounds and variability of lighting conditions are characteristics known to affect model performance in datasets. These results shed light on the importance of a well-prepared dataset in training a successful machine learning model and draw strategies regarding dataset augmentation to lead to improved robustness of the model. In conclusion, this work contributes to the field of mycology by providing a reliable, efficient, and accessible tool for mushroom classification. It also sets a precedent for future research into the application of lightweight neural networks for real-time and on-site biological classification tasks, promoting broader adoption of artificial intelligence driven identification systems in ecological and public health domains.