Detecting Drinks at a Crime Scene with Smart Cameras

Researchers used special cameras that can see many wavelengths of light to study how different drinks leave marks on surfaces. They set up a fake crime scene and collected images of nine types of beverage stains: papaya, coffee, pomegranate, orange, tea, wine, whisky, rum, and brandy. The camera captured 204 different color bands from visible to near‑infrared light. To focus on the most useful data, the team applied a statistical test called ANOVA. This step cut the number of bands down to 162, removing redundant information that could confuse the analysis. The cleaned data were then fed into four kinds of artificial‑intelligence models: a simple feed‑forward neural network, a one‑dimensional convolutional net, a long short‑term memory model, and a hybrid CNN‑LSTM network.Training involved adjusting the models’ learning rates automatically and stopping early when performance stopped improving. The researchers used a five‑fold cross‑validation method, ensuring each model was tested on diverse subsets of the data. This approach helped avoid overfitting and gave a realistic picture of each model’s accuracy. The results were impressive. The plain neural network (MLP) achieved the highest success rate, correctly identifying stains 95. 58% of the time. The other models also performed well, but none matched the MLP’s performance in this setting. These findings suggest that combining hyperspectral imaging with deep learning can reliably detect and classify drink stains without damaging the evidence. This technique could become a valuable tool for forensic teams looking to preserve crime scenes while gathering detailed information.