2.1 Sensor information fusion technology
Data layer fusion is to use multiple sensors to collect the information of the target to be measured, then directly fuse the original data, extract the target feature vector from the fusion information, and finally obtain the recognition results according to the feature vector information. In the data layer fusion, the sensor detects the same characteristics of the target to be measured, that is to say, the acquired data are homogeneous. Data layer fusion is not suitable for heterogeneous data, which is also the limitation of data layer fusion [6].
Data layer fusion is the fusion of the original data collected by sensors, that is, the fused data has the greatest authenticity and comprehensiveness. Therefore, the results of data layer fusion are more accurate than those of feature level fusion and decision level fusion. However, data layer fusion also has its own limitations, which are mainly reflected in the following aspects: data layer fusion faces a large number of original data, sometimes it can be said to be massive, which will inevitably increase the amount of calculation [7, 8]. The number of fusions is greatly increased. The fusion takes a long time, and the real-time performance of the system cannot be guaranteed. Since most of the original data detected by sensors are incomplete and uncertain, it is necessary for the fusion system to have good robustness and error correction ability to obtain high-precision target discrimination. Data layer fusion can only be performed for homogeneous data, but not for heterogeneous data. Due to the large amount of data detected by sensors, the fusion system cannot distinguish all interference information and cannot guarantee that each data is reliable. Therefore, it is necessary to enhance the anti-interference performance. Data layer fusion is usually used for multi-source image synthesis and direct synthesis of detection signal waveform. This fusion mode is also called pixel level fusion [9, 10]. Weight refers to the relative importance of a factor in the overall evaluation. Carrier frequency refers to the frequency used to load a signal onto a wave of a fixed frequency. Pulse width refers to the duration of the pulse that can reach the maximum value in the electronic field.
2.2 Biomedical testing
The biosensor is based on the change of signal generated by the reaction between the detected substance and the detection reagent. It has attracted people’s attention because of its fast, sensitive, accurate, and high selectivity in the process of biological detection. Several common biosensors are briefly introduced. The biosensor based on reflectance spectrum is the most widely studied and effective detection method. It is mainly based on the principle of thin film interference [11, 12]. When the refractive index of the substrate is affected by the analysis object, the reflection spectrum will move, so that the analyte can be detected quantitatively according to the reflection spectrum offset. The biosensor based on fluorescence spectrum makes use of the photophysical properties of the fluorescent group in the biosensor, since this is affected by the analysis target, the output form of the fluorescence signal changes, such as the shift and fluctuation of the fluorescence peak position. Fluorescent biosensors are a combination of research in photophysical chemistry and synthetic organic chemistry. Due to the high sensitivity of fluorescence analysis, this type of sensor has attracted people’s attention. Surface plasmon resonance biosensor has the advantages of no damage, no label, and real-time detection of biomolecule dynamic reaction process and high sensitivity. It has become a new technology with rapid development in recent years, and it is also one of the research hotspots in the application field of sensor detection [13]. SPR is a physical optical phenomenon. SPR sensor mainly uses the surface enrichment effect of metal nanoparticles and local surface plasmon resonance effect to improve the detection sensitivity of biosensor. As SPR is very sensitive to the refractive index of metal surface dielectric, the response strength of SPR will be different as long as the refractive index of surface dielectric changes.
2.3 Biological index detection method
The detection method used in this paper is to use the change of reflection spectrum signal of porous silicon as the sensing signal of biomolecule or chemical composition. In other words, the recognition and detection of biomolecules by porous optical biosensor is mainly based on the change of refractive index signal before and after the coupling of porous silicon and biomolecules. In the range of visible light or longer wavelength, the aperture of porous silicon layer is much smaller than the wavelength of incident light [11].
$$ \left(1-\rho \right)\frac{n_{si}^2-{n}_{eff}^2}{n_{si}^2+2{n}_{eff}^2}+\rho \frac{n_{air}^2-{n}_{eff}^2}{n_{si}^2+2{n}_{eff}^2}=0 $$
(1)
$$ f(x)=\frac{1}{Nh}{\sum}_{i=1}^Nk\left(\frac{X_i-x}{h}\right) $$
(2)
$$ k(x)=\frac{1}{\sqrt{2\pi }}\exp \left(-\frac{x^2}{2}\right) $$
(3)
When porous silicon is used as a sensor for detecting gas or solution, the gas or liquid diffuses into the pores of porous silicon [14]. When the volume fraction of gas or liquid in porous silicon is V (0 < V < ρ) and constant, the porous silicon layer will become a mixture of air, silicon, and gas or liquid molecules, and its effective refractive index satisfies the following conditions:
$$ \left(1-\rho \right)\frac{n_{si}^2-{n}_{eff}^2}{n_{si}^2+2{n}_{eff}^2}+\left(\rho -v\right)\frac{n_{air}^2-{n}_{eff}^2}{n_{air}^2+2{n}_{eff}^2}+v\frac{n_n^2-{n}_{eff}^2}{n_n^2+2{n}_{eff}^2}=0 $$
(4)
$$ {h}_t=\tanh \left({w}_c{x}_t+{u}_u\left({r}_t\Theta {h}_{t-1}\right)+{b}_c\right) $$
(5)
$$ {h}_t={z}_t\Theta {h}_{t-1}+\left(1-{z}_t\right)\Theta {h}_t $$
(6)
The film interference formula is obtained from the above formula:
$$ 2\mathrm{d}\left(\sqrt{{\mathrm{n}}_{\mathrm{eff}}^2-{n}_{air}^2}{\sin}^2i\right)= m\lambda $$
(7)
$$ 2{n}_{e_{ff}}d= m\lambda $$
(8)
$$ {\mathrm{E}}_{\mathrm{j}}\frac{\frac{1}{2{u}_j}{\sum}_{i=1}^{n_j}{\sum}_{r=1}^{n_j}\left|{y}_{ji}-{y}_{jr}\right|}{n_j^2} $$
(9)
It can be seen from the above formula that when the effective refractive index changes, the position of the interference peak will move [15, 16]. Therefore, the number of biomolecules entering the porous silicon layer can be determined by the relationship between the effective refractive index and the position shift of the interference peak, so as to realize the sensitivity calculation of the porous silicon biosensor and realize the detection of biological indicators.
2.4 Multi-sensor information fusion algorithm
2.4.1 Weighted average method
Weighted average method is a real-time processing fusion algorithm. Its essence is to process the information from multiple sensors in the future, and then weighted average the weight of a single sensor to obtain the final fusion result. This method is suitable for obtaining fusion value in dynamic environment, and the difficulty lies in solving the weight problem of single sensor [17]. Firstly, the algorithm analyzes the dataset and obtains the information of each dimension, including the span of each dimension. Analyze the recordset dataset to obtain the minimum and maximum values of each dimension in the recordset. In order to promote grid clustering, the maximum and minimum values need to be modified. The maximum and minimum values of each dimension are corrected according to the following formula.
$$ {\min}_i={\min}_i-k\times steplen(i) $$
(10)
$$ {\max}_i={\max}_i+k\times steplen(i) $$
(11)
$$ {u}_{\left(j/i\right)}={w}_{ij}{A}_i $$
(12)
$$ {s}_j=\sum \limits_i{c}_{ij}{u}_{\left(j/i\right)} $$
(13)
2.4.2 Bayesian reasoning
This method is suitable for data fusion in non-dynamic environment. It uses probability distribution to express information. In this method, each node in the sensor system is regarded as Bayesian estimation [18, 19]. When there is certain evidence (measure), Bayesian reasoning can determine the probability of hypothetical events according to the evidence (measurement); when there is new evidence (measure), the probability function of the assumed event can be updated by using the probability of new evidence and the likelihood function before generating new evidence. Finally, the system decision is given by some criteria. Bayesian network reasoning is a tool for decision support and causal discovery when information is incomplete. It is based on the probability distribution and considers that the values of all variables are controlled by the probability distribution [20]. Based on the observed data, the correct decisions can be made by calculating these probabilities. Because it provides a quantitative hypothesis method based on evidence support, it not only provides a theoretical basis for the algorithm of direct manipulation probability but also provides a theoretical framework for the analysis of algorithms without explicit probability calculation formula. Therefore, probability inference in Bayesian learning plays an important role in machine learning. The core of Bayesian network reasoning is to calculate the posterior conditional probability distribution [21, 22]. If the set of all variables is x, the set of evidence variables is e, and the set of query variables is Q, then the task of Bayesian network reasoning is to calculate the conditional probability distribution under the given set of evidence variables [23]. It can be formally described as:
$$ p\left(Q\left|E=e\right.\right)=\frac{p\left(Q,E=e\right)}{p\left(E=e\right)} $$
(14)
$$ y=\left( Da\times Db\right)-1 $$
(15)
$$ E=\frac{\sum_{j=1}^k{\sum}_{h=1}^k{\sum}_{t=1}^{n_j}{\sum}_{r=1}^{n_h}\left|{y}_{ij}-{y}_{hr}\right|}{2{n}^2u} $$
(16)
$$ Ew={\sum}_{j=1}^k{G}_{jj}{P}_j{s}_j $$
(17)
2.4.3 Clustering analysis
Clustering analysis is a heuristic algorithm, which mainly clusters data according to some similarity criteria, and decomposes each data group into corresponding target categories [24]. This method is widely used in target recognition, and the number of clusters is not clear.
2.5 Interoperable access control technology
For a large-scale distributed system, divide and conquer management is usually adopted, which leads to the concept of “domain,” that is, the system blocks constitute the “domain.” The traditional access control models, such as autonomous access control model, mandatory access control model, or role-based access control model, are mainly used in centralized systems and cannot meet the requirements of inter domain interoperability access control. For the collaborative work environment composed of multiple distributed and heterogeneous security domains (autonomous domains), multi domain security interoperability provides a secure and effective way to realize data interaction and resource sharing among domains.
Security domain is a bounded area composed of protected objects and user groups, and a set of security policies is managed and maintained by security administrators. Each domain can use different security models, classification patterns, syntax, and constraints to express its own information security policies, and has the right to allow legitimate requests and prevent illegal access. Security domain is also called autonomous domain.
Multi domain is composed of multiple security domains with mutual contract relationship. When a security domain needs to allow unknown entities or users to access local resources, corresponding security mechanisms must be set up to ensure that the access can be controlled within the scope of pre-defined known sharing policies. When such cross domain access is allowed between security domains, and each domain still uses the local original security policy, it can be regarded as a multi domain environment composed of these security domains:
$$ {\mathrm{V}}_{\mathrm{a}}={\sum}_{j=2}^k{\sum}_{h=1}^{j-1}{G}_{jh}\left({P}_j{s}_h+{P}_h{s}_j\right){D}_{jh} $$
(18)
$$ {y}_t={\sum}_{j=2}^k{\sum}_{h=1}^{j-1}{G}_{jh}\left({p}_j{s}_h+{p}_h{s}_j\right){D}_{jh}\left(1-{D}_{jh}\right) $$
(19)
There are two main characteristics of interoperable access control:
-
1)
Distributed management: every independent distributed application has its own security policy, so it is difficult to realize access control in a unified and centralized way:
$$ \ln \left(\frac{FI_{it}}{FI_{it}-1}\right)\ast {k}_{t1}\left[i\right]=\sum \limits_j\cos \left({w}_i^1,{w}_j^2\right)+\alpha +\beta \mathrm{In}\kern0.5em {FI}_{it}-1+\varphi {X}_{it}-1+{v}_i+{\tau}_t $$
(20)
$$ \theta =-\frac{1}{T}\mathrm{In}\left(1+\beta \right) $$
(21)
$$ {\sigma}_{\mathrm{ik}\mathrm{jl}}=\left\{\begin{array}{ll}\frac{\mathrm{n}}{\Delta_{\mathrm{ik}\mathrm{jl}}}\sqrt{\sum \limits_{\mathrm{s}=1}^{\mathrm{n}}{\left({\mathrm{x}}_{\mathrm{ik}}\left(\varepsilon \right)-{\mathrm{x}}_{\mathrm{jl}}\left(\varepsilon \right)\right)}^2{\Delta}_{\mathrm{ik}\mathrm{jl}}\left(\varepsilon \right)}& {\Delta}_{\mathrm{ik}\mathrm{jl}}>0;\\ {}0& {\Delta}_{\mathrm{ik}\mathrm{jl}}<0\end{array}\right. $$
(22)
$$ {\mathrm{X}}_1=\left\{{\mathrm{X}}_1\left(\mathrm{t}\hbox{-} 4\Delta \mathrm{t}\right),{\mathrm{X}}_1\left(\mathrm{t}\hbox{-} 3\Delta \mathrm{t}\right),{\mathrm{X}}_1\left(\mathrm{t}\hbox{-} 2\Delta \mathrm{t}\right),{\mathrm{X}}_1\left(\mathrm{t}\hbox{-} \Delta \mathrm{t}\right)\right\} $$
(23)
-
2)
Cross domain access control: if we want to achieve access control in distributed environment, we must cooperate among the security domains formed by independent distributed applications to decide whether to grant access to local resources to users in the external domain.
In multi-tier applications, the security of each domain becomes the primary problem of multi domain interoperability. Therefore, the access control of secure interoperability is mainly based on the following two principles:
-
1)
Autonomy principle: if an access is allowed in a single domain system, the access is also allowed in the case of multi domain interoperability:
$$ {\mathrm{Y}}_1=\left\{{\mathrm{Y}}_{1,1},{\mathrm{Y}}_{1,2},{\mathrm{Y}}_{1,3},{\mathrm{Y}}_{1,4},{\mathrm{Y}}_{1,5}\right\} $$
(24)
$$ {y}_i=\beta \left({u}_i,{v}_i\right)+\sum \limits_{j=1}^p{\beta}_j\left({u}_i,{v}_i\right){x}_{ij}+{\varepsilon}_j{\beta}_j $$
(25)
-
2)
Security principle: if an access is forbidden in a single domain system, the access is also prohibited in the case of multi domain interoperability. These two principles are formally described as follows:
A: U, R, P, and D represent user set, role set, permission set and security domain set of multi domain respectively;
B: UD, RD, and PD are recorded as u, R, P in security domain D, respectively;
C: d1, d2, d3,...dn represent the domains in the multi domain security interoperability.
So, \( \forall P\in {P}_{d_i}\left(i=1,2,\dots n\right)\Rightarrow P\in {P}_D,\kern0.5em \forall P\notin {P}_{d_i}\left(i=1,2,\dots n\right)\Rightarrow P\notin {P}_D\cdot \kern0.5em \)
$$ {\mathrm{Y}}_2=\left\{{\mathrm{Y}}_{2,1},\kern0.5em {\mathrm{Y}}_{2,2},\kern0.5em {\mathrm{Y}}_{2,3},{\mathrm{Y}}_{2,4},{\mathrm{Y}}_{2,5}\right\} $$
(26)
$$ {{\mathrm{w}}_{\mathrm{G}}}^{{\mathrm{A}}_{\mathrm{i}}{\mathrm{A}}_{\mathrm{j}}}=\max \left\{0,{W}_G\cdot \varepsilon \left({f_G}^{A_i},{f_G}^{A_j}\right)\right\} $$
(27)