Typically, LIBS systems consist of a few components that can be configured to meet specific requirements for scientific research, or they can be compact and rugged for field measurements in the field. Generally, these can be divided into (i) short solid-state pulses, (ii) Q-switched lasers, (iii) optics, (iv) detectors and spectrometers, and (v) computer systems. This system can provide a close-up analysis and a standalone configuration (Figure 1). In LIBS, a pulsed laser is used at high power on a sample surface (1), which causes the molecules of the sample to separate into their constituent atoms. A small amount of the sample is converted into vapor as a result of rapid energy discharge (2), resulting in the formation of high-temperature micro plasma accompanied by the characteristic sound of an ultrasonic shockwave (3) [13]. The plasma loses its energy as soon as it expands into space and begins to atomize. A spectrometer can detect the light emitted by ions and atoms during the relaxing stage. The sample is subjected to a laser pulse with a high peak power to cause a spark breakdown in the test sample medium. When a laser pulse is applied, the plasma reaches temperatures of up to 4000–15,000 K [13]. Molecules and particles within the sample are dissociated due to the energetic spark. Electrons and ions thereby excite the atoms. Cooling the plasma allows the excited ions and electrons to return to their ground states. Each emits wavelengths specific to its element and its fingerprint (signature). The composition can be determined by measuring the wavelengths and intensities of specific atomic emission lines (Figure 2).

Time-resolved LIBS measurements are the most common LIBS measurement used for improving signal-to-background ratios and minimizing the amount of continuum background generated during the initial plasma phase. Theoretically, plasma emission lines are analyzed under local thermodynamic equilibrium (LTE) conditions. The LTE can determine the plasma temperature and electron density, which can then be used to study ionization and atomization processes. Laser-induced plasma can vary significantly in size and shape depending on the ambient conditions, such as pressure, gas composition, and gas density [14]. Machine learning (ML) is considered a technique for learning from enormous amounts of data and extracting the unknown features for estimation or labeling. Typically, sample data are analyzed to determine whether there is a relationship between input and output data. It has generally been applied in emerging areas, especially in determining the authenticity of herbs. The complexity and diversity of the herb complicate LIBS spectral analysis, as well as a large number of factors. LIBS can extract the maximum amount of relevant data combined with machine learning techniques. This allows for greater accuracy in classification and quantitative analysis. Despite the volatility of sample spectra, LIBS coupled with machine learning can frequently offset this problem to a great extent. In recent years, ML has made significant advances beyond the traditional chemometric method [15]. It now provides robust and continuously updated algorithms and tools for developing spectroscopic data treatment models [16,17]. RF, for example [18], SVM [19], etc., have been utilized widely in quantitative and qualitative LIBS spectral analyses. Leo Breiman introduced RF in 2001, a regression technique based on multiple classifiers [18]. This strategy is based on the bootstrap method, which is applied continuously in creating test and training samples. RF generates several classification tree forms, and final prediction results are based on a simple majority vote of a single classification tree. In addition to being a useful technique for prediction, RF can also provide higher levels of accuracy. However, these algorithms suffer from a number of inherent shortcomings, which, in turn, has restricted their use in LIBS. Most traditional neural networks use gradient descent algorithms to update their parameters. Thus, the training process is slow, and there is a risk of falling into a local optimum. Thus, many new algorithms have been developed that are highly efficient in generalization at very fast learning rates to overcome these disadvantages. Kernel-based extreme learning machine (K-ELM) is a relatively new nonlinear method that has demonstrated excellent performance in classification and regression applications. As a result, a kernel function can replace the hidden layer of ELM and be used to eliminate the problems presented by traditional neural network training algorithms in terms of speed and over-fitting [20].

In plasma production, pulsed lasers are commonly used (LIBS). Specifically, laser characteristics are crucial to plasma production in gaseous, liquid, and solid media. Short-duration laser pulses with different wavelengths can generate multi-million-watt laser pulses from the infrared to the ultraviolet spectrum. Advanced lasers can produce vibrations that contain many billions to trillions of watts. In a fraction of a second, high-power laser pulses can vaporize metallic or refractory surfaces. Powerful lasers capable of delivering their energy to specific locations are significant. More importance is placed on the power per unit area provided to the target rather than its absolute value in LIBS. Depending on the laser wavelength, a power unit is called an “irradiance.” Conventional light sources with kilowatts of power are unable to focus as well as laser radiation and, therefore, cannot produce the same effects that lasers can [21].

3. Advantages and Disadvantages of LIBS

4. Prospects Related to Using LIBS in Herbal Technology