<![CDATA[The classification of the image of chemical solutions is crucial in laboratory automation and chemical industry applications; however, it remains challenging when solutions such as H2O with HCl and H2O with NaOH exhibit nearly identical visual characteristics under imaging conditions, particularly when their spectral fluctuation patterns are visually subtle. This study proposes an image classification framework that integrates Otsu-based segmentation in the HSV color space with convolutional neural network (CNN) models to classify High Height Fluctuation (HHF) images generated from a Multi-Scale chemical detection system (MSCS). The dataset consists of 102 HHF images, evenly distributed between the two solution classes. Transfer learning is applied using three CNN architectures, namely EfficientNetV2S, DenseNet201, and EfficientNetB0, and performance is evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis. The experimental results show that DenseNet201 achieves the best overall performance, while EfficientNetV2S provides competitive results with computational efficiency and Efficient-NetB0 yields a lighter model with lower recall. These findings indicate that combining segmentation with modern CNN architectures can effectively improve classification robustness in chemically similar solutions. This study presents a practical framework that combines Otsu-based HSV segmentation with transfer-learning CNNs to classify chemically similar solutions, providing actionable insights for deep learning-based chemical sensing applications.]]>
