Researchers have developed a novel technique which combines Raman spectroscopy and deep learning algorithms to allow for accurate on-site differentiation and identification of CA series synthetic cannabinoids. Given their low reactivity and subtle structural variations, on-site identification of the synthetic psychoactive substances has proved challenging with existing methods.
The research, published in the journal Analytical Chemistry, involved developing a convolutional neural network (CNN) algorithm which is integrated with an attention mechanism module to integrate with Raman spectroscopy to allow for precise differentiation of CA series synthetic cannabinoids. While Raman spectroscopy is used extensively for trace substance detection, the technique tends to fall short with structurally similar compounds due to similarities in their spectra.
While initial tests used only the CNN models, it was not until the team incorporated the SENet (Squeeze-and-Excitation Network) attention mechanism module that acceptable accuracy was obtained. Once integrated with the ResNet34 CNN model, the attention mechanism model elevated the classification accuracy of six synthetic cannabinoids to 100%.
In additional testing the newly developed SE_ResNet34 model proved accurate for generalized testing, and maintained its 100% accuracy irrespective to target concentration, structural analog presence, or common drug interferences.