Researchers at Zhengzhou Police University have developed an AI-powered Raman spectroscopy method that achieves 100% accuracy in identifying plastic beverage bottles.
Forensic analysis plays an important role in law enforcement. At crime scenes, forensic experts examine physical evidence to gather information that could lead to the apprehension of a suspect. Recently, a group of researchers at Zhengzhou Police University explored how to aid forensic investigations using spectroscopic techniques and artificial intelligence (AI). The study’s findings were published in the Journal of Raman Spectroscopy, and they demonstrated the utility of using convolutional neural networks (CNNs) and Raman spectroscopy for analyzing physical evidence (1).
In this study, the researchers examined how CNN and Raman spectroscopy can effectively analyze plastic beverage bottles, which are one of the more common physical objects found at crime scenes. The identification of these bottles can help provide important forensic evidence that would connect specific suspects to a specific crime (1). One of the current challenges forensic analysts face is that current methods used for plastic identification are not as accurate as required, and they also take time to perform (1). Because time is a valuable resource in criminal investigations, forensic analysts prefer using methods that can conduct the necessary analysis quickly.
Enter Raman spectroscopy, a technique routinely used for plastic identification (2). As an alternative method for plastic identification, Raman spectroscopy serves as a promising upgrade over traditional methods because it is non-destructive and uses laser light to determine molecular composition (1). By enhancing this technique with artificial intelligence, researchers have taken forensic plastic identification to a new level.
In this study, the research team used Raman spectroscopy and analyzed 40 commercially available plastic beverage bottles. The samples were initially categorized into two primary plastic types: polyethylene terephthalate (PET) and polyethylene (PE) (1). PET, the most common material for beverage bottles, was further divided into three subcategories using K-means clustering, an algorithm that groups data based on similarities (1).
Various data preprocessing techniques were used in the study to improve the accuracy of the spectral analysis. The researchers used the Savitzky–Golay (SG) algorithm for smoothing, standard normal variate (SNV), multiple scattering correction (MSC), and first-order derivatives, which helped reduce noise and standardize the data (1). The effect was that it allowed the researchers to use the machine learning models to classify materials easier (1).
Then, as per their experimental procedure, CNN was used to process the Raman spectral data. CNNs are widely used in image and pattern recognition tasks, making them ideal for detecting subtle spectral differences between plastic types (1,3). The performance of the CNN was evaluated using four key metrics: accuracy, precision, recall, and F1-score (1). Recall measures the proportion of actual positives correctly identified, while the F1-score balances precision and recall to provide a single performance metric.
When the CNN model was combined with SG and MSC preprocessing techniques, a perfect classification rate was achieved. This level of precision is a significant improvement over conventional classification methods and highlights the potential of AI-driven spectroscopy in forensic investigations (1).
Going forward, the research presented here is not only relevant for criminal investigations. Other application areas, such as environmental analysis, also benefit from the findings in this study, as monitoring plastic pollution is a
Given the growing role of AI in forensic science, Liu’s study represents a significant step toward more efficient, technology-driven investigations. With 100% classification accuracy, this study proves that AI-powered plastic identification is not just a theoretical possibility but a practical reality that could reshape forensic investigations worldwide (1).
