A port cargo real-time monitoring method based on sparse representation

Abstract

The application discloses a port cargo real-time monitoring intelligent scene adaptation method based on sparse representation, which comprises the following steps: arranging wireless signal transmitters and receivers outside the cargo box, the signal transmitters emit wireless signals of specific frequencies to the cargo box, and the signal receivers receive the signals transmitted out after the cargo, obtaining the quantitative data of the shielding effect of different types of cargos on radio waves and the labels of the cargo types, and through analyzing the characteristics of the received signals, the sparse representation technology is used to match the pre-established signal dictionary of the cargo types, so that the types of the cargos can be accurately identified. The application creates a signal feature dictionary representing different cargo types, each cargo type corresponds to a specific signal mode in the dictionary, the system compares the received signals with each signal mode in the dictionary, and determines the specific type of the cargo based on this. The method can efficiently classify cargos, eliminate manual opening and inspection in the traditional method, reduce the dependence on manual operation, and has high recognition accuracy and reliability, and is especially suitable for the logistics and transportation industries which need to automatically identify and classify a large number of cargos.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of real-time detection of port cargo transportation, and particularly relates to a port cargo transportation real-time monitoring method based on video signal sparse representation. BACKGROUND

[0002] With the development of today's logistics, as an important node of modern logistics system, the scale and throughput of the port are increasing, the types of goods are various, the transportation environment is complex and changeable, and the management and monitoring of port goods are difficult and complex. The traditional monitoring method relies on manual and mechanical detection, which is time-consuming and laborious and is easily affected by human factors when facing large-scale and high-density cargo transportation. Therefore, using radio technology to automatically monitor the goods in the container can eliminate the traditional manual opening and checking, which is beneficial to the accuracy of monitoring and the reduction of labor cost.

[0003] At present, common cargo identification technologies in the logistics industry include bar code identification, radio frequency identification (RFID) and image recognition. Among them, although the bar code identification and RFID technology improve the identification efficiency to a certain extent, they require the goods to carry corresponding labels. In actual application, it is more complex to realize. Although the image recognition technology can identify through monitoring and other visual information, in a complex transportation environment, factors such as light conditions, angle changes and goods stacking methods may affect the accuracy of the identification result.

[0004] Based on this background, the present application provides a cargo category identification method based on signal sparse representation, which obtains signal information of the goods passing through the container by arranging signal transmitters and receivers outside the container, and uses sparse representation technology to classify the goods in real time. This method not only improves the identification speed, but also reduces the dependence on manual operation, and has high practical value and application potential. SUMMARY

[0005] The purpose of the present application is to provide a port cargo real-time monitoring method based on sparse representation, which realizes the classification of goods through sparse representation analysis of wireless signals. The automatic level of port logistics is improved, and the monitoring cost is reduced.

[0006] To achieve the above purpose, the present application provides the following scheme: the present application provides a port cargo real-time monitoring method based on signal sparse representation, comprising the following steps:

[0007] (1)Signal transmitter and receiver are arranged outside the container. The signal transmitter transmits wireless signals of a specific frequency to the inside of the container, and the signal receiver receives the signals transmitted after passing through the goods. Quantitative data of the shielding effect of the target on the radio waves in the case where the target is loaded with different types of goods and the label data of the target goods category are obtained, and the data are divided to obtain a training sample set and a test sample set; wherein the training sample set includes the received radio data of the target on different goods categories and the position label of the target;

[0008] (2) The quantitative data in all cases are processed, and the data of each type of goods are processed into a column vector. All column vectors obtained are spliced in the form of columns to form a dictionary, and a signal feature dictionary containing different goods categories is created. Each goods category corresponds to a specific signal pattern in the dictionary;

[0009] (3) The target equation is constructed, and the linear fitting error of the test signal and the regularization term are considered for iterative solution to obtain the final sparse vector;

[0010] (4) Joint analysis and calculation of several maximum values in the sparse vector are performed to obtain the predicted category of the target.

[0011] The method of obtaining label data and dividing data in step (1) comprises the following steps:

[0012] 1) According to the actual situation, select the possible object categories, each category corresponds to a unique digital label;

[0013] 2) A plurality of radio receiving sensors are arranged around a monitoring area in a uniform distribution, the positions are fixed, and a radio transmitting sensor is configured, which has K variable transmitting positions. The wireless communication between a transmitting sensor and a receiving sensor is referred to as a link, and there are N=K*L links in the entire experimental scene;

[0014] 3) When the target container arrives at the detection site, the radio transmitting sensor sends radio signals, and all receiving sensors receive the signals. The transmitting sensor moves all the selectable positions and transmits radio signals, and all receiving nodes receive the signals at the same time. Then, the signal strength is extracted from the spectrum of each link as effective information, and an information matrix with a dimension of L rows and K columns is obtained for each container detection. At the same time, the digital label information of the target container loaded with various goods is saved, which corresponds to the true category information one by one;

[0015] 4) The collected data set is divided into a training sample set and a test sample set at a certain ratio (such as 5:1).

[0016] The splicing method of the column vector in step (2) comprises the following steps:

[0017] Take a class of goods to box, launch radio signal test, N receivers get signal, convert the obtained information matrix into a column vector with dimension N, and mark the information as 1; repeat the measurement τ times, and after conversion into column vectors, a total of τ column vectors are obtained, and the label information is Lab1={1,1,…,1}; repeat the above two steps until the data of the goods categories required for constructing the dictionary are collected; respectively, the column vectors at each position are sequentially spliced to form the final dictionary, and each column of the dictionary is normalized by two norms, and the corresponding label information is spliced to form the annotation information.

[0018] The step (3) comprises the following steps:

[0019] Construct the target equation:

[0020]

[0021] The above target equation contains two parts of fitting error and regularization term. Wherein, y represents a test signal, W is a dictionary matrix, x is a sparse vector to be solved, represents the introduced non-convex GMC regularization; α, λ are positive parameters; z is an auxiliary vector;

[0022] In the process of solving practical problems, since the number of categories of goods existing in a certain container is far less than the total number of goods categories, a sparse coding method is used to solve the sparse vector.

[0023] Step (4) comprises the following steps:

[0024] 1) First, after the sparse vector x={x 1,1 ,x 1,2 ,…,x 1,τ ,x 2,1 ,...x S,τ} is solved, x∈R n ; summing the x coefficients of each sparse vector, that is, wherein 1≤p≤S, thus obtaining x * ={x1,...,x p ,...x S}, x * In order to obtain the weight vector corresponding to different labels, the greater the weight, the greater the probability of predicting the current position, and the elements in x * are arranged in descending order;

[0025] 2) Extract several maximum values in the weight vector, and each weight corresponds to a different goods category label;

[0026] 3) several weights are normalized, and a joint calculation is performed to obtain the predicted target class.

[0027] The normalization processing mode and the method for predicting the target class in step 3) are as follows:

[0028] Normalization processing: first, several maximum value vectors are normalized according to The normalization processing mode and the method for predicting the target class in step 3) are as follows:

[0029] Predicting the target class:

[0030]

[0031] Wherein, class represents the final target class, a i represents the normalized weight corresponding to position i, and class(i) represents the real class label corresponding to i.

[0032] The following technical effects are disclosed in the present application:

[0033] The method can quickly process a large number of goods, reduces the dependence on manual operation through the sparse representation technology of wireless signals, and improves the efficiency of goods classification. When modeling, the method of sparse coding is used to represent the goods category model, noise is processed by minimum fitting error in the target equation, non-convex regularization is introduced to ensure the sparsity of the sparse vector obtained, and the convexity of the target equation is maintained, which is convenient for fast solving. The wireless signal is not affected by the light condition and can be stably transmitted in various environmental conditions. The method can work effectively under different goods stacking modes and transportation environments, and has good adaptability. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0035] Figure 1 It is a schematic diagram of the port goods real-time monitoring model based on sparse representation of the present application.

[0036] Figure 2 It is a schematic diagram of the port goods real-time monitoring method based on sparse representation in the embodiment of the present application. DETAILED DESCRIPTION

[0037] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below, obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0038] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0039] The present application provides a port cargo real-time monitoring method based on sparse representation, as shown in Figure 2 The present application provides a port cargo real-time monitoring method based on sparse representation, as shown in

[0040] The actual cargo classification problem is modeled, and based on the fact that the cargo to be detected in the actual problem is much less than the total number of possible cargo categories, the problem is converted into a sparse coding problem during modeling. First, all cargo category test signals are collected through an external radio sensor to form a dictionary. In this process, the test signals collected can be obtained by sparse coding to obtain a sparse vector containing classification information, and finally a joint analysis calculation method is used to predict the cargo category.

[0041] (1) A signal transmitter and a receiver are arranged outside the cargo box. The signal transmitter transmits a wireless signal of a specific frequency to the inside of the cargo box, and the signal receiver receives the signal transmitted out after passing through the cargo. Quantitative data of the shielding effect of the radio wave of the target in the case where the target is loaded with different categories of goods and label data of the target cargo category are obtained, and the data is divided to obtain a training sample set and a test sample set; wherein the training sample set includes the received radio data of the target on different cargo categories and the position label of the target;

[0042] (2) Process the quantitative data under all conditions, process the data of each cargo into a column vector, and splice all the column vectors to form a dictionary to create a signal feature dictionary containing different cargo categories. Each cargo category corresponds to a specific signal pattern in the dictionary;

[0043] (3) Construct the target equation, simultaneously consider the linear fitting error of the test signal and the regularization term, and iteratively solve to obtain the final sparse vector;

[0044] (4) Jointly analyze and calculate several maximum values in the sparse vector to obtain the target prediction category.

[0045] The method for obtaining label data and dividing data in step (1) comprises the following steps:

[0046] 1) According to the actual situation, select the possible object categories, each category corresponds to a unique digital label;

[0047] 2) As Figure 1 , a uniform distribution of L radio receiving sensors is arranged around a monitoring area, the position is fixed, a radio transmitting sensor is configured, which has K variable transmitting positions, the wireless communication between a transmitting sensor and a receiving sensor is called a group of links, and there are N = K * L groups of links in the entire experimental scene;

[0048] 3) When the target box arrives at the detection site, the radio transmitting sensor sends a radio signal, and all receiving sensors receive the signal; the transmitting sensor moves all the selectable positions and transmits the radio signal, and all receiving nodes receive the arrival signal at the same time; then, the signal strength is extracted from the spectrum of each link as valid information, and an information matrix with a dimension of L rows and K columns is obtained for each box detection; at the same time, the digital label information of the target box under various goods conditions is saved, which corresponds to the true category information one by one;

[0049] 4) The collected data set is divided into a training sample set and a test sample set at a ratio of 5:1.

[0050] The splicing method of the column vector in step (2) comprises the following steps:

[0051] Take a type of goods to box, transmit a radio signal test, N receivers get the signal, convert the obtained information matrix into a column vector with a dimension of N, and mark the label information as 1; repeat the measurement τ times, respectively convert into column vectors, and get τ column vectors, and the label information is Lab1 = {1, 1, …, 1}; repeat the above two steps until the data of the goods categories required for constructing the dictionary are collected; respectively, the column vectors at each position are sequentially spliced to form the final dictionary, and each column of the dictionary is normalized by the two-norm, and the corresponding label information is spliced to form the annotation information.

[0052] The step (3) comprises the following steps:

[0053] Construct the target equation:

[0054]

[0055] The above target equation contains two parts of fitting error and regularization term. Wherein, y represents a test signal, W is a dictionary matrix, x is a sparse vector to be solved, represents the introduced non-convex GMC regularization; α, λ are positive parameters; z is an auxiliary vector;

[0056] In solving the actual problem, since the number of product categories in a certain container is always much smaller than the total number of product categories, a sparse coding method is used to solve the sparse vector.

[0057] Step (4) includes the following steps:

[0058] 1) After obtaining the sparse vector x = {x 1,1 ,x 1,2 ,…,x 1,τ ,x 2,1 ,...x S,τ}, x ∈ R n , sum the x coefficients of each sparse vector, i.e. where 1 ≤ p ≤ S, to obtain x * = {x1,...,x p ,...x S}, x * To obtain the weight vector corresponding to different labels, the greater the weight, the greater the probability of predicting the current position, and arrange the elements in x * in descending order;

[0059] 2) Extract several maximum values in the weight vector, and each weight corresponds to a different product category label;

[0060] 3) Normalize the several weights and jointly calculate to obtain the predicted target category.

[0061] The normalization processing method and the method for predicting the target category in step 3) are:

[0062] Normalization processing: first normalize the several maximum value vectors according to , where num is the number of maximum values, i.e. the normalized weight;

[0063] Prediction of target category:

[0064]

[0065] where class represents the final target category, a i represents the normalized weight corresponding to position i, and class(i) represents the real category label corresponding to category i.

[0066] Finally, it should be noted that the above-described embodiments are merely specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, but not to limit the same. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that any skilled person familiar with the technical field can still modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features therein; and these modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. All should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for real-time monitoring of port cargo based on sparse representation, characterized in that: The method comprises the following steps: (1) obtaining quantization data of the shielding effect of radio waves and label data of the categories of goods in a case where the case contains different categories of goods, and dividing the obtained data to obtain a training sample set and a test sample set; wherein the training sample set comprises radio data received by cases containing different categories of goods and category labels of the goods; (2) processing the quantization data of all categories of goods, processing the data of each category into a column vector, and forming a dictionary by splicing all column vectors obtained above according to columns; (3) constructing an objective equation, and iteratively solving the linear fitting error of a test signal and a regularization term to obtain a final sparse vector, specifically comprising the following steps: 1) Construct the objective function: The objective equation comprises a fitting error and a regularization term, wherein y represents a test signal, W is a dictionary matrix, x is a sparse vector to be solved, represents a non-convex GMC regularization term introduced, and α and λ are positive parameters; z is an auxiliary vector; 2) In the process of solving actual problems, the number of categories of goods in a case is much smaller than the total number of categories of goods, so a sparse coding method is used to solve the sparse vector; (4) jointly analyzing and calculating several maximum values in the sparse vector to obtain a to-be-detected category of goods, and the joint analysis and calculation process comprises the following steps: 1) Obtain the sparse vector x = {x p,i}, where x ∈ R n , x p,i represents the sparse coefficient corresponding to the pth category of goods and its ith measurement result, satisfying 1≤p≤S, 1≤i≤τ, τ is the number of repeated measurements of each category of goods, and S is the total number of categories of goods; sum the sparse coefficients of each category of goods under all measurements to calculate the weight x p of each category of goods: Further, a complete weight vector is obtained as follows: x * = {x1, x2,..., x S} 2) extracting several maximum values in the weight vector, each weight corresponding to a different category label of goods; 3) normalizing the several weights, and jointly calculating to obtain a predicted target category, and the calculation steps are as follows: First, normalize the several maximum value vectors according to to obtain the normalized weight; then, the target category is predicted as follows: wherein class denotes the class of the final target, a i denotes the normalized weight corresponding to position i, and class(i) denotes the real class label corresponding to class i.

2. The port goods real-time monitoring method based on sparse representation according to claim 1, characterized in that: (1) the goods are classified by category, and each category corresponds to a unique digital label; (2) L radio receiving sensors are evenly distributed outside the case, the positions are fixed, a radio transmitting sensor is configured, K kinds of goods are selected as experimental objects, and there are N=K*L kinds of possible signal data in the entire experimental scene; (3) the category of the goods is unknown, a radio transmitting sensor is used to send a radio signal, and all receiving sensors receive the signal; experiments are performed on multiple categories of goods, a radio signal is transmitted, and all receiving nodes simultaneously receive the arriving signal; then, the signal strength is extracted from the spectrum of each link as effective information, and each kind of goods obtains an information matrix with a dimension of L rows and K columns; meanwhile, digital label information representing the category of the goods is saved, which corresponds to the information represented by the goods in one-to-one correspondence; (4) the collected data set is divided into a training sample set and a test sample set at a preset ratio.

3. The method of claim 1, wherein, The column splicing method of the column vector is: Take a class of goods to box, launch radio signal test, N receiver gets signal, will get information matrix into a dimension size N column vector, will its label information as 1;Repeat measurement τ times, respectively into column vector after a total of τ column vector, label information Lab1={1,1,…,1};Repeat the above two steps until the data needed to collect the dictionary construction goods category;Respectively, the column vector in each position is sequentially spliced to form the final dictionary, and the label information corresponding to each column of the dictionary is spliced to form the label information.

4. The method according to claim 3, wherein, The two-norm normalization method of the column vector m is: where m = [m1, m2,..., m n ] T is the column vector, m i (1≤i≤n) represents the i-th component of the column vector, and n is the dimension of the column vector; 5. The method of claim 1, wherein, The solving of the sparse coding adopts a forward-backward splitting algorithm, the forward splitting improves the sparse representation by gradually adding new basis vectors, and the backward splitting improves the coding quality by optimizing the existing basis vector set;The two steps are alternately performed until the convergence condition is met or the predetermined stopping criterion is reached.

Images