Bayesian network spam classification method based on multi-source information fusion

Abstract

The application discloses a kind of multi-source information fusion's bayesian network garbage classification method, comprising the following steps: step one: using image sensor and other at least one type of sensor from garbage sample corresponding feature information is extracted, obtains the initial data of garbage sample;Initial data is preprocessed using Laplace smoothing method, and training sample is obtained;Step two: with image sensor as main and combined with the multi-source heterogeneous feature information of other type sensor, multi-source information fusion model is established using bayesian parameter estimation algorithm;Step three: according to the connection between the characteristic information of garbage sample and garbage category, bayesian network is established using training sample and bayesian classifier is constructed, multi-source information fusion model is simplified using bayesian classifier, and multi-source information fusion simplified model is obtained;Step four: test sample is input into multi-source information fusion simplified model, and according to the calculation result of maximum posteriori estimation value, the category of the garbage sample is judged.

Description

Technical Field

[0001] This invention belongs to the field of waste sorting technology, specifically a Bayesian network waste sorting method based on multi-source information fusion. Background Technology

[0002] Waste sorting is an effective measure to protect the ecological environment and promote economic development. At present, waste sorting technology has become a research hotspot in academia and industry, and some achievements have been made. However, there are still some limitations in the field of special waste sorting.

[0003] Bayesian parameter estimation methods utilize Bayes' theorem, combined with new evidence and historical prior probabilities, to obtain new probabilistic expressions. This method proposes a way to calculate the probability of a hypothesis based on the prior probability of the hypothesis, the probability of observing different data under a given hypothesis, and the observed data itself. Bayesian estimation treats the parameter to be estimated as a random variable conforming to a certain prior probability distribution, observes the sample, and then uses mathematical analysis to derive the prior probability distribution into a posterior probability distribution. Finally, it uses information from the sample to correct the initial estimate of the parameter.

[0004] Multi-source information fusion methods combine multiple (similar or dissimilar) information sources in space or time according to a specific standard to provide a linear interpretation or description of the object under test, thereby improving the performance of the information system. This system's fusion model adopts a distributed fusion system architecture and performs information fusion processing at the decision-making level. It fully utilizes multi-dimensional information sources and combines redundant or complementary information from multiple sources in space or time according to a set configuration to obtain a linear interpretation or description of the object under test. This results in the information system exhibiting better performance than a system composed of its constituent subsets, ultimately leading to a joint decision result. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a Bayesian network garbage classification method based on multi-source information fusion, which can improve the discrimination accuracy of garbage samples, especially garbage samples with complex feature information.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A Bayesian network-based garbage classification method based on multi-source information fusion includes the following steps:

[0008] Step 1: Extract relevant feature information from the garbage samples using an image sensor and at least one other type of sensor to obtain the initial data of the garbage samples; preprocess the initial data using the Laplace smoothing method to obtain training samples;

[0009] Step 2: Using image sensors as the main source and combining multi-source heterogeneous feature information from other types of sensors, establish a multi-source information fusion model using the Bayesian parameter estimation algorithm;

[0010] Step 3: Based on the relationship between the feature information of the garbage samples and the garbage categories, a Bayesian network is built using the training samples, and a Bayesian classifier is constructed using the Bayesian network. The Bayesian classifier is then used to simplify the multi-source information fusion model, resulting in a simplified multi-source information fusion model.

[0011] Step 4: Extract relevant feature information from the waste samples to be classified, and preprocess the initial data of the waste samples to be classified using the Laplace smoothing method to obtain test samples; input the test samples into the multi-source information fusion simplification model to obtain simplified multi-source fusion information, and determine the category of the waste sample based on the calculation result of the maximum a posteriori estimate.

[0012] Furthermore, in step one, the feature information of the waste sample includes image, odor, and material. Image information of the waste sample is collected using an image sensor, odor information of the waste sample is collected using an odor sensor, and material information of the waste sample is collected using a material sensor.

[0013] Furthermore, the method for preprocessing the initial data using the Laplace smoothing method is as follows:

[0014] Given sample set information , For a set of feature attributes, If the set is categorical information, then the conditional probability of categorical information in the overall sample is:

[0015]

[0016] The conditional probability of a feature attribute in its classification information is:

[0017]

[0018] in, Representation of features The Each attribute value; Representing feature information The One feature attribute; Indicates the first in the category kind; Representation of features The total number, that is ; Indicates the smoothing coefficient; For threshold function; Indicates the total number of waste samples; Indicates the quantity of different types of waste; Indicates the sample set number The attribution labels of each sample are , Represents the sample's first Dimensional feature attributes; Indicates sample No. The value of the feature attribute is the first... indivual.

[0019] Furthermore, in step two, the expression for the multi-source information fusion model is:

[0020]

[0021] in, This represents the optimal estimate of the maximum posterior probability after fusing the feature information extracted by n types of sensors. ={ , , ,..., } represents the feature information of garbage samples extracted by n types of sensors. Indicates the first Feature information of garbage samples extracted by sensor-like devices; ={ , , ,..., } represents the category information of the garbage sample. Indicating garbage samples Information by category, ; Indicate category The decision estimate.

[0022] Furthermore, in step three, the simplified expression for the multi-source information fusion model is:

[0023]

[0024]

[0025] in, This represents the conditional probability formula in probability theory, specifically meaning that when the type of waste is... When, the feature attribute is The probability of; The specific practical meaning is that when the type of waste is When, the feature attribute is The probability of; express Prior probability of garbage class; The total number of sensors, This represents the value of the positive signal sensor.

[0026] The beneficial effects of this invention are as follows:

[0027] This invention presents a Bayesian network-based waste classification method using multi-source information fusion. First, it employs Laplace smoothing to preprocess initial data, obtaining training and test samples to address the zero prior probability problem inherent in traditional Bayesian methods, eliminating the influence of zero-valued prior probabilities on the fusion result. Then, based primarily on image sensor judgments, it combines multi-source heterogeneous feature information from multiple other types of sensors, utilizing a Bayesian parameter estimation algorithm to establish a multi-source information fusion model. Building upon the Bayesian network and classifier constructed based on the relationship between waste sample features and waste categories, the multi-source information fusion model is simplified to obtain a simplified model. Finally, test samples extracted from the waste samples to be classified are input into the simplified model, and the discrimination result is obtained by calculating the maximum a posteriori probability estimate. Experimental verification demonstrates that this invention's Bayesian network-based waste classification method can improve the discrimination accuracy of waste samples, especially those with complex feature information, proving that it can reduce the high ambiguity in the discrimination process between hazardous and recyclable waste, thereby obtaining more accurate classification results. This has important theoretical significance and practical value for the classification of complex waste in daily life. Attached Figure Description

[0028] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration:

[0029] Figure 1 This is a flowchart of the Bayesian network garbage classification method based on multi-source information fusion of the present invention;

[0030] Figure 2 This is a schematic diagram of a Bayesian network;

[0031] Figure 3 This is a Bayesian network structure diagram for multi-feature information and garbage categories. Detailed Implementation

[0032] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0033] like Figure 1 As shown in the figure, the Bayesian network garbage classification method based on source information fusion in this embodiment includes the following steps.

[0034] Step 1: Extract relevant feature information from the garbage samples using an image sensor and at least one other type of sensor to obtain the initial data of the garbage samples; preprocess the initial data using the Laplace smoothing method to obtain training samples.

[0035] This embodiment establishes the following waste sorting principles:

[0036] (1) Recyclable material Y1 refers to unpolluted waste generated in daily life that is suitable for recycling and can be utilized as a resource.

[0037] (2) Perishable waste Y2 refers to food scraps generated in farmers' markets and residents' daily lives.

[0038] (3) Hazardous waste Y3 refers to waste that poses a direct or potential threat to human health or the natural environment.

[0039] (4) Other waste Y4 refers to household waste other than perishable waste, recyclables and hazardous waste.

[0040] This embodiment pre-defines three types of sensors and several types of multi-source heterogeneous data. The characteristic information of the waste sample includes image, odor, and material. An image sensor collects image information of the waste sample, an odor sensor collects odor information, and a material sensor collects material information. The image sensor can effectively identify four types of waste, but it may misidentify waste with complex characteristics due to insufficient utilization of that information. The odor sensor can identify hazardous waste (volatile organic compounds), such as CH3OH, SO2, CH4, and H2S. The material sensor can identify hazardous waste (batteries) and recyclable waste (non-ferrous metals), such as Hg, Ni, Pb, Au, Ag, Cu, Fe, and Al. Ordinary samples refer to waste samples that the image sensor can correctly identify, such as mineral water bottles; special samples refer to waste samples identified by auxiliary sensors such as odor sensors, such as mineral water bottles containing pesticides.

[0041] To calculate the prior probability data required for Bayesian estimation, ensure the efficient performance of the classifier, and address the zero-probability problem caused by insufficient training data, this embodiment employs the Laplace smoothing method to preprocess the initial data. Specifically, the method for preprocessing the initial data using Laplace smoothing is as follows:

[0042] Given sample set information , For a set of feature attributes, If the set is categorical information, then the conditional probability of categorical information in the overall sample is:

[0043]

[0044] The conditional probability of a feature attribute in its classification information is:

[0045]

[0046] in, Representation of features The Each attribute value; Representing feature information The One feature attribute; Indicates the first in the category kind; Representation of features The total number, that is ; This represents the smoothing coefficient, which is typically set to 1. For threshold function, =1; Indicates the total number of waste samples; Indicates the quantity of different types of waste; Indicates the sample set number The attribution labels of each sample are , Represents the sample's first Dimensional feature attributes; Indicates sample No. The value of the feature attribute is the first... indivual.

[0047] Step 2: Using image sensor data as the primary source and incorporating heterogeneous feature information from other sensor types, a multi-source information fusion model is established using a Bayesian parameter estimation algorithm. Specifically, the process of establishing the multi-source information fusion model is as follows.

[0048] The essence of Bayesian estimation is to obtain the optimal estimate of parameters through Bayesian decision-making, thereby minimizing the total expected risk. Suppose the tested waste sample has m categories. ={ , , ,..., The characteristic information of n types of sensors is ={ , , ,..., Multi-feature decision fusion algorithms estimate the decision value by combining the feature information values ​​of n types of sensors according to a certain estimation criterion function (gas density range, material type detection, image feature information, etc.). The true value is used to represent it. When the feature information of a certain sensor feature is... Then, the category The decision estimate is set as The loss function is According to the Bayesian estimation criterion, it is defined in the sample The following are the conditional risks:

[0049]

[0050] in, Indicate category The posterior probability; This represents the probability distribution of the sample data.

[0051] because If the condition is non-negative and the risk function is minimized, then the Bayesian estimation of the decision can be obtained:

[0052]

[0053] Based on the Bayesian estimation decision above, the optimal estimate of the maximum posterior probability of a single sensor feature information is:

[0054]

[0055] in, This represents the optimal estimate of the maximum posterior probability of a single sensor feature.

[0056] In the Naive Bayes parameter estimation algorithm, the feature information of each sensor is independent of each other. By generalizing from the feature information of a single sensor to the feature information of multiple sensors, the expression of the multi-source information fusion model is obtained as follows:

[0057]

[0058] in, This represents the optimal estimate of the maximum posterior probability after fusing the feature information extracted by n types of sensors. ={ , , ,..., } represents the feature information of garbage samples extracted by n types of sensors. Indicates the first Feature information of garbage samples extracted by sensor-like devices; ={ , , ,..., } represents the category information of the garbage sample. Indicating garbage samples Information by category, ; Indicate category The decision estimate.

[0059] Step 3: Based on the relationship between the feature information of the garbage samples and the garbage categories, a Bayesian network is established using training samples, and a Bayesian classifier is constructed using the Bayesian network. The Bayesian classifier is then used to simplify the multi-source information fusion model, resulting in a simplified multi-source information fusion model.

[0060] 3.1 Bayesian Networks

[0061] A Bayesian network is a directed acyclic graph that represents the probability distributions and relationships between variables. ,in It is an acyclic directed graph (DAG), a qualitative description of the dependencies between features, consisting of a set of nodes and directed edges from parent nodes to child nodes; node set The nodes in the graph correspond one-to-one with the variables in the actual problem; directed edge set This represents the dependencies between feature attributes; the presence or absence of directed edges indicates whether the features are conditionally independent; conditional probability distribution set. It is a set of parameters describing the conditional probabilities of network layout, quantitatively representing the dependency relationship between each feature and its parent node, where... ∈ This indicates that the Bayesian network at a given node When the parent node The conditional probability.

[0062] Assume the set of random variables in the practical problem ,in This corresponds to each node in a Bayesian network. The steps for constructing a Bayesian network are:

[0063] (1) Establish the correspondence between garbage classification categories, characteristic attributes and network nodes based on actual problems.

[0064] (2) Based on the dependency or causal relationship between variables, a directed acyclic graph representing conditional independence is established. The joint probability of all nodes represented by the graph can be expressed as the product of the conditional probabilities of each node. If we use... Representing variables The set of parent nodes, then The joint probability distribution is:

[0065]

[0066] in, Representing variables The set of parent nodes.

[0067] (3) Determine the set of parameters for the conditional probability of the Bayesian network layout.

[0068] 3.2 Bayesian Network Construction Process

[0069] Given a set of features , A set of sensor feature information, This is a garbage category set. A Bayesian network is constructed based on the steps of building a Bayesian network. The conditional probabilities of each node are as follows: Figure 2 As shown. Constructing the Bayesian network also yields the conditional probability of each node in the network, which is:

[0070] .

[0071] in, Represents the prior probability of the type of waste; This indicates that, given the specific type of waste, the waste's characteristic attributes are: The probability of; This indicates that, given the specific type of waste, the waste's characteristic attributes are: The probability of; This indicates that, given the specific type of waste, the waste's characteristic attributes are: The probability of; This indicates that one of the waste types and characteristic attributes is Given this information, the characteristic attribute of this waste is: The probability of; This indicates that one of the waste types and characteristic attributes is Given this information, the characteristic attribute of this waste is: The probability of.

[0072] 3.3 Simplifying the Multi-Source Information Fusion Model Using a Bayesian Classifier

[0073] Given sensor feature information for detecting waste samples The probability of garbage is:

[0074]

[0075] in, Denotes the regularization factor, and express Prior probability of garbage class; This indicates that, given the specific type of waste, the waste's characteristic attributes are: The probability of.

[0076] Each feature information node in the constructed Bayesian network is defined. and waste categories Related, and sensor feature information If they are independent of each other, the above formula can be expressed as:

[0077]

[0078] Therefore, the expression of the multi-source information fusion model can be rewritten as:

[0079]

[0080] Regularization factor Given fixed values ​​for the sensor feature information input, the maximum a posteriori estimation problem can be simplified to a maximum likelihood estimation problem, i.e.:

[0081]

[0082] In waste sorting technology, image detection methods yield higher accuracy results. Therefore, if the image sensor is set as the primary sensor, then:

[0083]

[0084] in, { } represent recyclable waste, hazardous waste, perishable waste and other waste categories respectively; The decision value is based on sensor feature information, and the output is... This indicates that the waste sample's test result is unknown / recyclable / perishable / hazardous / other waste, and the test value has reached the sensor's set threshold. If the sensor outputs a positive characteristic value, then the sensor outputs a positive characteristic value. The total number of sensors, This represents the value of the positive signal sensor.

[0085] This embodiment assumes that the classification of waste samples with complex features has special characteristics, and supplements the identification of hazardous waste and recyclable waste. Improvements and optimizations are made when calculating the posterior estimates of these two types of waste. The optimization considers the influence of parameters such as the total number of sensors, the number of features, the conditional probability of a single sensor, and the type of waste. To improve the accuracy of the model in classifying waste samples with complex features, a variable weighting factor is proposed.

[0086]

[0087] The final simplified expression for the multi-source information fusion model is:

[0088]

[0089] in, This represents the conditional probability formula in probability theory, specifically meaning that when the type of waste is... When, the feature attribute is The probability of; This indicates that the specific practical meaning is when the type of waste is When, the feature attribute is The probability of; express Prior probability of garbage class; The total number of sensors, This represents the value of the positive signal sensor.

[0090] After calculating the maximum a posteriori estimate using the above formula, the corresponding... The class represents the classification result obtained after the method judges the garbage sample.

[0091] Step 4: Extract relevant feature information from the waste samples to be classified, and preprocess the initial data of the waste samples to be classified using the Laplace smoothing method to obtain test samples; input the test samples into the multi-source information fusion simplification model to obtain simplified multi-source fusion information, and determine the category of the waste sample based on the calculation result of the maximum a posteriori estimate.

[0092] The following describes the specific implementation and technical effects of the Bayesian network garbage classification method based on multi-source information fusion in this embodiment, using concrete examples.

[0093] 1. Training sample data

[0094] Referring to the waste sorting principles established in this embodiment, 900 units of recyclable waste, 2000 units of perishable waste, 800 units of hazardous waste, and 1300 units of other waste were collected, and their feature information was extracted through relevant sensors. Based on the data in the training sample feature information table, combined with the training sample data preprocessing methods (12) and (13), and taking... The prior probability table is calculated as shown in Table 1.

[0095] Table 1 Feature Information Table

[0096]

[0097] (Continued from the table above)

[0098]

[0099] 2. Standard test samples

[0100] Information on the characteristics of routine waste in daily life was collected. The results of the collection of relevant information by the set image sensor, odor sensor and material sensor are shown in Table 2.

[0101] Table 2. Common Waste Samples and Their Characteristics (Partial)

[0102]

[0103] 3. Special test samples

[0104] The difficulty in sorting some types of waste lies in the high complexity of the feature information. In order to reflect the representativeness and effectiveness of the experiment, a special waste sample composed of complex feature information was established. The relevant feature information is shown in Table 3.

[0105] Table 3. Special Waste Samples and Their Characteristics (Partial)

[0106]

[0107] 4. Experimental Analysis

[0108] The experiment used three types of heterogeneous sensor data: an image sensor, an odor sensor (which can identify chemical components such as CH3OH, SO2, CH4, and H2S), and a material sensor (which can identify metals such as Hg, Ni, Pb, Au, Ag, Cu, Fe, and Al). These sensors were used to extract multi-source feature information from both regular and special samples. A Bayesian network waste classification method based on the fusion of multi-source information before and after the improvement was then used for comparative experiments, and the results were analyzed.

[0109] 4.1 Constructing Bayesian Networks

[0110] First, a Bayesian parameter estimation algorithm is used for multi-feature information fusion; second, using methods such as... Figure 3 The constructed Bayesian network is used to build a Bayesian classifier and simplify the information fusion results. Finally, the maximum a posteriori estimate is calculated using test samples to test the category of the waste samples to be classified.

[0111] 4.2 Experiment 1

[0112] Experiment 1 uses a fusion algorithm of Naive Bayes parameter estimation to combine regular garbage samples and garbage samples with complex feature information.

[0113] According to the classification principles, 100 units of each of the four categories of special waste in Table 2 were selected for a total of 4 × 50 = 400 units of test samples for numerical testing. The discrimination process of the Bayesian network waste classification method using multi-source information fusion was then implemented, including hazardous waste (… The process for judging the pesticide-containing plastic bottles {1,3,0,0,0,0,0,0,0,0,0,0,0} in the test sample is as follows:

[0114] Based on the constructed Bayesian classifier, and by substituting the sample data, we get:

[0115]

[0116]

[0117]

[0118]

[0119] because =1, therefore we can get From this, the posterior probability value can be obtained. , , , Based on the data calculation results, the maximum a posteriori estimate is thus obtained. The value was 0.996, thus classifying the waste sample as hazardous waste, a result different from the waste category of the test sample. The same steps were used to calculate the classification results for the remaining test samples, as shown in Table 4.

[0120] Table 4 Results of garbage sample discrimination based on complex feature information in Experiment 1

[0121]

[0122] According to the classification principles, 100 units of each of the four categories of waste in Table 1 were selected as test samples, totaling 4 × 50 = 400 units, for calculation and testing. The discrimination process of the Bayesian network waste classification method using multi-source information fusion was performed, and the discrimination results of the regular waste test samples were obtained through the same calculation steps as above, as shown in Table 5.

[0123] Table 5. Results of discrimination of routine waste samples in Experiment 1

[0124]

[0125] 4.3 Experiment 2

[0126] The experiment employed a waste identification and classification method based on Bayesian parameter estimation combined with multi-source information, and conducted experiments using both conventional waste samples and waste samples with complex feature information.

[0127] According to the classification principles, 100 units of each of the four categories of special waste in Table 2 were selected for a total of 4 × 100 = 400 units of test samples for numerical testing. The discrimination process of the improved Bayesian network waste classification method based on multi-source information fusion was then implemented, where hazardous waste = The process for identifying the pesticide-containing plastic bottles {1,3,0,0,0,0,0,0,0,0,0,0,0,} in the test samples is as follows:

[0128] From the simplified model of multi-source information fusion, we can obtain:

[0129]

[0130]

[0131]

[0132] The improved Bayesian network garbage classification method based on multi-source information fusion was used to calculate the garbage classification result. ~ The posterior estimate:

[0133]

[0134]

[0135]

[0136]

[0137] Based on the data calculation results, the maximum a posteriori estimate is obtained as follows: 4.400 Therefore, the waste sample was determined to be hazardous waste, and the determination result was the same as that of the test sample. Following the same steps, the determination results for the remaining test samples are shown in Table 6.

[0138] Table 6 Results of garbage sample discrimination based on complex feature information in Experiment 2

[0139]

[0140] According to the classification principle, 100 units of each of the four categories of waste in the conventional waste test samples in Table 2 were selected, totaling 4 × 100 = 400 units, for the calculation test. The discrimination process of the improved Bayesian network waste classification method using multi-source information fusion was performed, following the same calculation steps as above, and the discrimination results of the conventional waste test samples are shown in Table 7.

[0141] Table 7 Results of discrimination of routine waste samples in Experiment 2

[0142]

[0143] 4.4 Analysis of Experimental Results

[0144] Experiment 1 tested a fusion algorithm based on Naive Bayes parameter estimation on garbage samples with both conventional and complex features. The results showed an accuracy of 93.5% for conventional garbage samples and 89.5% for garbage samples with complex features. Experiment 2 tested a garbage identification and classification method based on a combination of Bayesian parameter estimation and multi-source information. The results showed an accuracy of 98.75% for conventional garbage samples and 98.5% for garbage samples with complex features.

[0145] Under the same training and test datasets, the improved Bayesian network waste classification method using multi-source information fusion can improve classification accuracy and robustness, with an overall improvement of 9% in classification accuracy. Specifically, the accuracy for distinguishing important recyclable waste and hazardous waste improved by 14%, and the maximum difference in accuracy among different waste category classification applications was only 4%. Therefore, it can be concluded that the improved Bayesian network waste classification method using multi-source information fusion is highly effective in classifying waste with complex features.

[0146] 5. Conclusion

[0147] To address the challenges of multi-source information processing in the complex waste sorting process, this paper proposes a waste identification and classification method based on the combination of Bayesian parameter estimation and multi-source information. Experiments demonstrate that the combination of these two methods can fully utilize the complex features of the waste to be classified, thereby improving the accuracy of classification.

[0148] Based on the definition of multi-source feature information fusion, and according to the inherent characteristics and treatment importance of recyclable and hazardous waste, an improved Bayesian network waste classification method based on multi-source information fusion is proposed, which assigns variable weight factors to the sensor feature information of waste samples.

[0149] Experimental results show that this method can accurately distinguish waste samples with complex features, and can set variable weight factors for sensor features of the model according to the characteristics of waste in different fields. Therefore, it can be extended to fields such as medical waste classification, industrial waste classification, and construction waste classification, and has a very broad application prospect and great application value.

[0150] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A Bayesian network-based garbage classification method based on multi-source information fusion, characterized in that: Includes the following steps: Step 1: Extract relevant feature information from the garbage samples using an image sensor and at least one other type of sensor to obtain the initial data of the garbage samples; preprocess the initial data using the Laplace smoothing method to obtain training samples; Step 2: Using image sensors as the main source and combining multi-source heterogeneous feature information from other types of sensors, establish a multi-source information fusion model using the Bayesian parameter estimation algorithm; Step 3: Based on the relationship between the feature information of the garbage samples and the garbage categories, a Bayesian network is built using the training samples, and a Bayesian classifier is constructed using the Bayesian network. The Bayesian classifier is then used to simplify the multi-source information fusion model, resulting in a simplified multi-source information fusion model. Step 4: Extract relevant feature information from the waste samples to be classified, and use the Laplace smoothing method to preprocess the initial data of the waste samples to be classified to obtain test samples; The test sample is input into the multi-source information fusion simplification model to obtain the simplified multi-source fusion information. The category of the garbage sample is determined based on the calculation result of the maximum a posteriori estimate. In step two, the expression for the multi-source information fusion model is: in, This represents the optimal estimate of the maximum posterior probability after fusing the feature information extracted by n types of sensors. ={ , , ,..., } represents the feature information of garbage samples extracted by n types of sensors. Indicates the first Feature information of garbage samples extracted by sensor-like devices; ={ , , ,..., } represents the category information of the garbage sample. Indicating garbage samples Information by category, ; Indicate category Decision estimate; In step three, the simplified expression for the multi-source information fusion model is: in, This represents the conditional probability formula in probability theory, specifically meaning that when the type of waste is... When, the feature attribute is The probability of; This indicates that the specific practical meaning is when the type of waste is When, the feature attribute is The probability of; express Prior probability of garbage samples; The total number of sensors, The value is a positive signal from the sensor. It is a variable weighting factor.

2. The Bayesian network garbage classification method based on multi-source information fusion according to claim 1, characterized in that: In step one, the feature information of the waste sample includes image, odor and material. Image information of the waste sample is collected using an image sensor, odor information of the waste sample is collected using an odor sensor, and material information of the waste sample is collected using a material sensor.

3. The Bayesian network garbage classification method based on multi-source information fusion according to claim 1, characterized in that: The method for preprocessing initial data using the Laplace smoothing method is as follows: Given sample set information , For a set of feature attributes, If the set is categorical information, then the conditional probability of categorical information in the overall sample is: The conditional probability of a feature attribute in its classification information is: in, Representation of features The Each attribute value; Representing feature information The One feature attribute; Indicates the first in the category kind; Representation of features The total number, that is ; Indicates the smoothing coefficient; For threshold function; Indicates the total number of waste samples; Indicates the quantity of different types of waste; Indicates the sample set number The attribution labels of each sample are , Represents the sample's first Dimensional feature attributes; Indicates sample No. The value of the feature attribute is the first... indivual.

Images