You can manage bookmarks using lists, please log in to your user account for this.
Media type:
E-Book
Title:
A Preliminary Study of Geographic Traceability of Soil Physical Evidence
:
Machine Learning Recognition of Elemental Fingerprints and Morphological Features
Description:
Soil is a type of physical evidence that is often found at crime scenes, but it is not efficiently utilized in case detection processes. In this research, soil samples collected from 20 soil sites scattered in five cities of China were investigated to compare the values and conditions of their application. Soil physical evidence traceability was analyzed from two dimensions, namely, soil composition and morphology. The elemental fingerprints of the samples were determined using inductively coupled plasma mass spectrometry (ICP-MS), and the samples were source classified by combining principal component analysis, partial least squares discriminant analysis, and support vector machine models. Images of the soil in the visible band were collected using a hyperspectral imaging system, and the samples were source classified by combining deep learning models, including BPNN and CNN. Good classification results were obtained for both soil traceability methods based on both ideas. Among them, the ICP-MS method combined with the PLS model was able to achieve 100% classification accuracy for trace soil samples, but the experimental cost was high and the pre-processing process was complicated. Meanwhile, the hyperspectral method combined with the CNN model was able to achieve 99.19% fast and nondestructive identification, but the demand for soil detection was high. Notably, the two traceability methods can be applied in different circumstances. The selection of a suitable analytical detection method for practical applications will facilitate the effective use of on-site soil physical evidence for the resolution of cases and attainment of breakthroughs in difficult cases