An intelligent garbage detection method based on light source visual lens recognition

Abstract

The application relates to the technical field of garbage classification processing, and discloses an intelligent garbage detection method based on light source visual lens recognition, which comprises the following steps: S1, capturing garbage and environmental image data through a visual sensor; S2, collecting images by using a quantum optics enhanced visual lens system; S3, converting the image data into a low-dimensional matrix by using a CANDECOMP / PARAFAC method; S4, inputting the low-dimensional matrix into a deep neural network for garbage recognition and classification; and S5, calculating the robot path by using a path optimization algorithm to complete a garbage grabbing task. Through the cooperative action of a quantum superposition state light source and a quantum entangled state light source, the garbage image data is subjected to enhanced processing, so that the definition and detail resolution of the image are improved, the interference of environmental noise is effectively reduced, the details and edges of the garbage object are more obvious, and the garbage classification accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of waste sorting and treatment technology, specifically to an intelligent waste detection method based on light source vision lens recognition. Background Technology

[0002] Current waste sorting methods typically rely on computer vision and traditional machine learning algorithms to identify and classify waste through image processing techniques. These methods generally use ordinary visual sensors or cameras to acquire environmental images and rely on image enhancement and feature extraction for classification. Traditional methods depend on manual feature selection or the use of classic image processing techniques such as edge detection and color extraction, which have improved the accuracy of waste sorting to some extent. Meanwhile, some more advanced waste sorting systems have begun to utilize deep learning, especially convolutional neural networks, to automatically learn the features of waste objects, thereby improving classification accuracy. However, despite these technological advancements, existing technologies still face some limitations that affect the overall effectiveness of the system.

[0003] First, ordinary visual sensors cannot produce image quality sufficient for accurate classification in complex environments, especially in low-light and high-noise scenes, where image details are often blurred, affecting the accuracy of garbage object identification. Second, while existing machine learning and deep learning methods have improved classification accuracy to some extent, they still face problems of low computational efficiency and long training times, particularly when processing large-scale garbage images, where the training process is very slow and difficult to adapt to the needs of real-time classification. Furthermore, the application of existing path planning algorithms in garbage collection robots mainly relies on static maps, lacking real-time adaptability to dynamic environments, causing the robots to be unable to effectively cope with obstacles and environmental changes. These problems limit the further development of automation and intelligence in garbage sorting systems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent waste detection method based on light source vision lens recognition, which solves the problems of insufficient accuracy and automation level in existing waste detection methods.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent waste detection method based on light source vision lens recognition, comprising the following steps:

[0006] S1. Capture image data of waste and the environment through a visual sensor;

[0007] S2. Acquire the image data using a quantum optics-enhanced vision lens system;

[0008] S3, tensor decomposition is performed on the acquired image data, and a CANDECOMP / PARAFAC decomposition method is used to convert the multi-dimensional image data into a low-dimensional matrix, to obtain a compressed data representation;

[0009] S4, the converted low-dimensional matrix is input into a deep neural network for training of a classification model, to identify garbage objects in the image first, and then to perform feature extraction and classification;

[0010] S5, according to the environmental image data, a path optimization algorithm is used to calculate the robot path, so that the robot can complete the garbage grabbing task in a changing environment.

[0011] Preferably, the environmental image includes at least one dimension of data selected from an RGB image, a depth image, or multi-dimensional image data;

[0012] The quantum optics enhanced visual lens system includes a quantum superposition state light source and a quantum entangled state light source, and by adjusting the phase and amplitude of the light source, an enhanced light field distribution in the image acquisition process is formed.

[0013] Preferably, the tensor decomposition is realized by a high-order tensor decomposition method, and a CANDECOMP / PARAFAC decomposition is used;

[0014] The CANDECOMP / PARAFAC decomposition method extracts the main spatial features and texture information in the image by a step-by-step optimization method, and stores them in the form of a low-dimensional matrix, to reduce data redundancy.

[0015] Preferably, the tensor decomposition step includes:

[0016] The image data is organized into a high-dimensional tensor, and each dimension in the tensor represents a different attribute or spatial feature of the image;

[0017] The CANDECOMP / PARAFAC decomposition is applied to the high-dimensional tensor, and the tensor is decomposed into a series of low-dimensional matrices, each of which represents an important feature of the original data;

[0018] The low-dimensional matrix is further processed by a deep neural network after decomposition, to ensure that the garbage objects in the image are effectively identified and classified.

[0019] Preferably, the path optimization algorithm is based on the variational method and the steepest descent method for path optimization, and the steps of path optimization include:

[0020] On the basis of considering the environmental map and obstacles, a kinematic model of the robot is established, and the variational method is used to optimize the path to calculate the best driving trajectory;

[0021] The path is adjusted in real time by using the steepest descent method, the path is corrected according to the change of dynamic environment data, and the robot can efficiently reach the target position to pick up the garbage.

[0022] Preferably, the environment data in the path optimization includes the current position of the robot, the position of the obstacle in the environment and the expected target position.

[0023] The path optimization updates the environment data in real time, optimizes the cost function of the path by combining the variational method, and corrects the path by combining the steepest descent method, so that the robot can perform the garbage picking task in an unknown or dynamically changing environment.

[0024] Preferably, the training of the classification model adopts a quantum annealing method.

[0025] The training process includes optimizing the objective function by the quantum annealing method.

[0026] The quantum annealing method improves the classification performance of the garbage classification model by adjusting the parameters of the quantum state.

[0027] Preferably, the objective function in the quantum optimization algorithm includes a classification loss function and a regularization term.

[0028] The objective function aims to minimize the classification error, and adjusts the parameters by quantum optimization to optimize the accuracy of garbage classification.

[0029] Preferably, the training process of the classification model is executed by a quantum computing platform.

[0030] The quantum computing platform is configured with a quantum processing unit responsible for quantum optimization of model parameters.

[0031] The quantum optimization process is realized by a quantum annealing method, which mainly adjusts the weights and biases of the model.

[0032] Preferably, the image data is collected by a quantum optical enhanced visual lens system, which adjusts according to the quantum characteristics of the light source, enhances the micro details in the collected image, and makes the garbage features in the image more obvious.

[0033] The present application provides an intelligent garbage detection method based on light source visual lens recognition.

[0034] 1、The present application realizes the enhancement processing of garbage image data through the cooperative action of quantum superposition state light source and quantum entangled state light source, thereby improving the image definition and detail resolution. Compared with the traditional optical imaging system, the present application can effectively reduce the interference of environmental noise, make the details and edges of garbage objects more obvious, greatly improve the precision of the garbage classification system, provide clearer and richer input data for subsequent deep learning classification, and promote the improvement of garbage classification precision.

[0035] 2、The present application can convert image data into a low-dimensional matrix while retaining key feature information through high-order tensor decomposition technology for dimension reduction processing of image data. Compared with traditional dimension reduction technology, the tensor decomposition method of the present application effectively reduces redundant data, simplifies the computational complexity of subsequent deep neural networks, improves the precision of the garbage classification system, and is superior in garbage object recognition and classification in complex scenes, thereby providing technical support for realizing automatic and efficient garbage classification.

[0036] 3、The present application introduces a quantum annealing optimization algorithm in the training process of the garbage classification model, which can accelerate the training process of the garbage classification model and optimize the classification precision of the model. Compared with the traditional gradient descent-based optimization method, quantum optimization can find the global optimal solution faster, reduce the training time, improve the automatic decision-making ability of the model in complex environments, and further promote the automation level of the intelligent garbage classification system.

[0037] 4、The present application realizes path planning and dynamic adjustment of the robot by combining variational method and steepest descent method in robot path optimization. Compared with traditional path planning methods, the path optimized by combining variational method can dynamically adapt to changing environments, and the real-time adjustment by combining steepest descent method enables the robot to efficiently and accurately complete garbage grabbing tasks in complex environments, greatly improving the performance and efficiency of the robot in the process of automatic garbage collection and improving the overall automation level. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0040] Please refer to the drawings of the present application Figure 1The embodiment of the present application provides a kind of intelligent garbage detection method based on light source visual lens identification, comprising the following steps:

[0041] S1, garbage and environmental image data are captured by visual sensor;

[0042] Specifically, in the present embodiment, step S1 is mainly used to obtain garbage image data, which is the environmental image captured by visual sensor.The environmental image as input data is crucial for subsequent garbage recognition and classification process.The main purpose of this step is to ensure that the image data has sufficient quality to enable effective garbage object detection in subsequent image processing and analysis.

[0043] Generally, visual sensor adopts camera or other optical sensing device, can include infrared camera, RGB camera or depth camera etc., is used to obtain different kinds of environmental image.Specifically, the image data that visual sensor can capture should cover the multidimensional information of environment, such as the spatial features, color features and depth information of image etc.These information has important significance for the accuracy of garbage detection and classification.

[0044] In a possible implementation manner, visual sensor can capture three-dimensional image data in the environment, including depth scanning of objects in the environment, and the obtained image data includes pixel point information and depth coordinate information of image.Combination of this multidimensional information makes garbage detection not only rely on the content of two-dimensional image, but also can further improve the recognition accuracy of garbage object through depth information.

[0045] Further, the visual sensor in the present embodiment can combine certain preprocessing steps for signal enhancement when acquiring image data.Suppose the captured environmental image data is image matrix I (x, y), wherein x and y represent pixel coordinates in image respectively, and I represents the brightness value or color value of the pixel point.The processing of image data can be optimized using the following image enhancement formula:

[0046] I ′ (x,y)=α·I(x,y)+β;

[0047] Wherein, I ′ (x,y) represents the enhanced image matrix;Alpha is contrast adjustment coefficient;Beta is brightness adjustment constant.Through appropriately adjusting the value of alpha and beta, the details and contrast in image can be enhanced, and the image quality is improved, so as to subsequent garbage recognition.

[0048] After image data acquisition, further processing steps may involve image denoising and edge enhancement, which also require algorithmic optimization. For example, Gaussian filtering or median filtering can be used to remove noise from the image, resulting in smoother image data.

[0049] As an alternative, visual sensors can capture continuous environmental image data in real time, forming a dynamic image stream for subsequent real-time waste detection. By continuously acquiring environmental images, the position and state of waste objects can be updated in a timely manner in dynamic environments, thereby further enhancing the real-time performance and accuracy of waste identification.

[0050] In this embodiment, the visual sensor can work in conjunction with other sensor systems to form a multi-sensor fusion system. For example, sensor fusion technology can combine data from different sensors (such as temperature sensors, pressure sensors, etc.) to provide more comprehensive environmental information, thereby providing more accurate data support for subsequent waste object detection.

[0051] Through the above-mentioned technical means, the environmental image data acquired by the visual sensor in this embodiment can effectively support subsequent waste identification and classification, especially under complex or dynamic environmental conditions, to ensure the reliability and accuracy of waste detection results.

[0052] S2. Use a quantum optics-enhanced vision lens system to acquire image data;

[0053] Specifically, in this embodiment, step S2 involves using a quantum optics-enhanced vision lens system to acquire and enhance the environmental image data obtained in step S1. The core objective of this step is to utilize the properties of quantum optics to optimize the quality of the captured image by adjusting the phase and amplitude of the light source, thereby ensuring the accuracy of subsequent waste detection and sorting.

[0054] Generally, quantum optics-enhanced vision lens systems include quantum superposition light sources and quantum entangled light sources. Quantum superposition light sources can generate multiple possible light states, while quantum entangled light sources produce a stronger optical interference effect by entangled multiple light waves. In this way, the light field output by the light source can enhance coherence during image acquisition, improving image detail. Specifically, by adjusting the phase and amplitude, the light source can effectively increase the coherence of light waves, significantly improving the signal-to-noise ratio of image acquisition.

[0055] For example, the process of acquiring image data can be described by the following formula:

[0056] I ′ (x,y)=η·I(x,y)·cos(φ);

[0057] Among them, I′ (x,y) represents the image data enhanced by the quantum optics system; I(x,y) is the original image data; η is the light source gain factor; φ is the light source phase difference; (x,y) is the image pixel coordinate. In this formula, cos(φ) describes the influence of the light source phase difference on the image enhancement effect. By precisely controlling the phase difference, the clarity of the image can be improved, especially in low-illumination or high-contrast scenes, the enhancement effect of the light field is particularly obvious.

[0058] In some embodiments, the light source of the quantum optics enhanced visual lens system can adjust the image light field using quantum interference effects. Through the phase interference between quantum light sources, the texture and edge details in the image are enhanced, especially the recognizability of small objects is improved. Through this technology, the system can always maintain the clarity and high quality of the image under different environmental light conditions.

[0059] As an option, the quantum optics enhanced visual lens system can combine an autofocus technology to adjust the focal length in real time according to the captured image information. This autofocus can ensure that the garbage objects are accurately and clearly captured at any depth. For example, in some cases, the image may contain multiple levels of objects, and through the autofocus technology, the system can dynamically focus on the target garbage object, further improving the recognition accuracy.

[0060] In a possible implementation, the quantum optics system can use a laser as a light source. The laser has the characteristics of strong coherence and stable wavelength, and can generate highly consistent light waves. By adjusting the output beam frequency and phase of the laser, the light source can have a highly concentrated light field, thereby improving the resolution of the image. For example, by adjusting the beam intensity and phase difference of the light source, clear image data can be obtained in a low-light environment, ensuring that the recognition of garbage objects is not disturbed by the environment in different scenarios.

[0061] In addition, in some embodiments, the quantum optics enhanced visual lens system can further reduce image distortion caused by atmospheric interference or lens distortion through adaptive optics technology. By quickly adjusting the optical elements in the lens, the system can obtain more stable and accurate image data in a dynamic environment.

[0062] In summary, the quantum optics enhanced visual lens system in this embodiment can not only improve image quality, but also ensure that the system can always obtain image data with high resolution and clarity under different environmental conditions, laying a solid foundation for subsequent garbage recognition and classification.

[0063] S3, performing tensor decomposition on the obtained image data, using CANDECOMP / PARAFAC decomposition method to convert the multi-dimensional image data into low-dimensional matrices, obtaining the compressed data representation.

[0064] Specifically, in the present embodiment, step S3 involves performing tensor decomposition processing on the image data obtained from step S2. The purpose of this step is to convert high-dimensional image data into low-dimensional matrix representation, thereby simplifying the subsequent image feature extraction and classification process. This process provides a computationally convenient simplified data representation for subsequent deep neural network processing.

[0065] Generally, image data contains a large amount of pixel information, and directly processing these high-dimensional data will consume a large amount of computing resources and may cause performance bottlenecks. Therefore, reducing the dimensionality of image data through tensor decomposition method can effectively reduce the redundancy of data and extract the key features of the image. Specifically, tensor decomposition technology can convert multi-dimensional image data into a set of low-dimensional matrices through the decomposition of high-order tensors, and these matrices can contain the main information of the image and have good representation ability.

[0066] In the present embodiment, CANDECOMP / PARAFAC decomposition method is used to perform tensor decomposition on image data. This method represents image data as a high-order tensor and decomposes it into multiple low-dimensional matrices, each of which represents a part of the features of the image data. The goal of CANDECOMP / PARAFAC decomposition is to find a set of factor matrices such that their product can approximate the original high-order tensor.

[0067] Specifically, image data is first organized into a three-dimensional or higher-dimensional tensor, each dimension of the tensor representing an attribute or feature of the image, such as color, spatial coordinates, brightness, etc. Assume that the tensor of the original image data is:

[0068] where I, J and K are the three dimensions of the tensor, representing different features of the image; denotes the set of real numbers.

[0069] Then, by applying the CANDECOMP / PARAFAC decomposition method, the tensor is decomposed into several factor matrices:

[0070]

[0071] where R is the rank of the decomposition, representing the dimension of the low-dimensional matrix after decomposition. The goal of this decomposition process is to minimize the following objective function:

[0072]

[0073] where a r , b r , c r are the r-th columns of matrices A, B, C, respectively; denotes the outer product operation of tensors; ∥.∥ F denotes the Frobenius norm, which measures the error between the decomposition result and the original tensor.

[0074] In one possible implementation, after CANDECOMP / PARAFAC decomposition, the low-dimensional matrices obtained contain the main features of the image data, such as texture, edge, and other key information. Through this process, the originally complex image data is simplified into a set of smaller-dimensional matrix representations, facilitating subsequent deep learning networks for feature extraction and classification.

[0075] As an option, the low-dimensional matrices obtained after tensor decomposition can be directly input into a deep neural network for further feature learning. In some embodiments, the structure of the deep neural network includes multiple convolutional layers, pooling layers, and fully connected layers, which can effectively extract high-level features of the image from the decomposed low-dimensional matrices for garbage classification and detection.

[0076] In further implementations, to improve computational efficiency, the low-dimensional matrices can be quantized to reduce computational load and improve processing speed. Specifically, in the representation of low-dimensional matrices, some unimportant elements can be compressed through quantization, thereby achieving efficient storage and processing of data.

[0077] In this embodiment, the image data is processed by tensor decomposition for dimensionality reduction, which effectively reduces the complexity of the data while preserving the main features of the image. This provides a simplified yet efficient data representation for subsequent garbage object detection. This technology can be effectively applied in various environments to improve the accuracy and processing speed of garbage detection.

[0078] S4, input the transformed low-dimensional matrices into a deep neural network to train the classification model, first identify the garbage objects in the image, and then perform feature extraction and classification;

[0079] Specifically, in this embodiment, step S4 involves feature extraction on the image data after tensor decomposition in step S3. The core goal of this process is to obtain key features that can represent the content of the image through specific mathematical transformations and statistical analysis. This step not only provides basic data for subsequent garbage recognition and classification, but also ensures the efficiency and robustness of the calculation.

[0080] Generally, the way of feature extraction is selected according to the dimension and data distribution of the image data. Since the data has been mapped to a low-dimensional space after tensor decomposition, methods such as principal component analysis (PCA), independent component analysis (ICA), or non-negative matrix factorization (NMF) can be used to further extract features. Specifically, the goal is to extract the most discriminative features to facilitate the training and inference of subsequent classification models.

[0081] In this embodiment, the low-dimensional matrices A, B, and B after tensor decomposition are first normalized to eliminate the influence of data scale difference on the feature extraction process. Assuming that the normalized form of matrix A is A ′ The calculation method is as follows:

[0082]

[0083] where μ A and σ A represent the mean and standard deviation of the elements of matrix A, respectively; A ′ ij is the element at position (i, j) in the normalized matrix A. Similar normalization processes are also applicable to matrices B and C.

[0084] As an option, key features can be further extracted based on singular value decomposition (SVD). The SVD method can further decompose the low-dimensional matrix to obtain the most important feature vectors. In some embodiments, matrix A ′ After SVD decomposition, it can be represented as:

[0085]

[0086] where U A and V A are the left and right singular matrices, respectively; S A is a diagonal matrix whose diagonal elements represent singular values; represents the transpose of matrix V A , i.e., interchanging the rows and columns of V A to obtain These singular values represent the main amount of information of the data and can be used for feature selection.

[0087] Specifically, when performing feature selection, a threshold τ can be set to retain only the feature components that satisfy the following condition:

[0088] s i ≥ τ · max(S A );

[0089] where s i is the s Athe i-th singular value of S; τ is an empirically set weight parameter; max(S A ) is the maximum value in the set S A . For example, in some cases, the top 95% cumulative energy contribution is selected, which can ensure data compression while maximizing information retention.

[0090] In one possible implementation, Gabor filters can also be used to perform spatial frequency analysis on image data to obtain local features such as edges and textures. Suppose the response function of a Gabor filter is:

[0091]

[0092] where (x ′ ,y ′ ) are the coordinates after rotation transformation; x and y represent the coordinates in the image; 2π is a constant representing a multiple of the value of pi; θ is the filter direction; λ is the wavelength; σ controls the scale; γ is the spatial aspect ratio parameter; G(x,y,θ,λ) is a filter representing some image processing, especially in processing image textures and edge detection. By adjusting these parameters, feature response values of different scales and directions can be obtained.

[0093] In some embodiments, to further improve the robustness of feature extraction, deep learning-based feature extraction methods such as convolutional neural networks (CNN) can be combined. In this case, the data after tensor decomposition can be used as input to extract deep features through multiple convolutional layers and pooling layers.

[0094] Generally, the feature map F l of the l-th layer of a convolutional neural network can be represented as:

[0095] F l = f(W l *F l-1 +b l );

[0096] where W l is the convolution kernel weight; b l is the bias term; f(.) is a nonlinear activation function; * represents convolution operation; F l-1 is the feature map of the previous layer.

[0097] Through the above method, the extracted features can be used as input to the garbage recognition model. The dimension of the feature vector is related to the task requirements, and both fixed-length features and dynamic feature selection combined with attention mechanisms can be used to improve the accuracy of garbage classification.

[0098] S5, calculate the robot path through a path optimization algorithm based on the environment image data captured by the visual sensor, the path optimization algorithm uses variational method to optimize the path, and combines the steepest descent method to dynamically adjust the path, so that the robot can complete the garbage grabbing task in the changing environment;

[0099] Specifically, in the present embodiment, step S5 involves classifying and recognizing the image using the features extracted in the previous steps. The goal of this step is to classify the garbage image and determine its category by constructing a suitable classification model. The accuracy of classification and recognition directly affects the performance of the system, so the selection and parameter adjustment of the classification algorithm are crucial in this step.

[0100] Generally, the classification process will select a suitable algorithm according to the extracted features and task requirements. In some application scenarios, traditional machine learning methods such as support vector machines (SVM), k-nearest neighbors (k-NN), and decision trees can be used as candidate algorithms. On the other hand, in recent years, deep learning methods such as convolutional neural networks (CNN) and long short-term memory networks (LSTM) have been widely used in image classification tasks due to their strong feature learning capabilities.

[0101] In the present embodiment, a convolutional neural network (CNN) is used for garbage classification. Specifically, at the beginning of step S5, the features extracted from step S4 are first input into the CNN network. In some embodiments, the feature map is convolved by the convolution layer and down-sampled by the pooling layer, thereby gradually extracting high-level abstract features. Assuming that at the l-th layer, the result of the convolution operation is F l , which can be represented as the l-th layer feature map F l of the convolutional neural network in step S4.

[0102] As an option, after convolution, the pooling layer reduces the dimension of the feature map, usually using maximum pooling MaxPool(F l ) or average pooling AveragePool(F l ). For example, assuming that the feature map after pooling is F p , which can be represented as:

[0103] F p = MaxPool(F l );

[0104] In one possible implementation, the size and step length of the pooling operation are adjusted according to task requirements, which directly affects the performance of the model.

[0105] Then, the pooled feature maps are further processed through several fully connected layers, and the probability distribution of each garbage class is obtained through the Softmax activation function. Specifically, the output vector y of the last fully connected layer i can be represented as:

[0106]

[0107] where z i is the i-th component of the output; C is the number of classes; exp(.) represents the exponential function; the exponential function is used to convert the output of the network into the probability distribution of each class.

[0108] In some embodiments, the classification model can further employ transfer learning techniques to improve the adaptability and generalization ability of the model on different datasets. In this case, transfer learning is performed using a pre-trained CNN model such as VGG16, ResNet, etc., which can accelerate the training process and effectively reduce the overfitting phenomenon.

[0109] Generally, in order to improve the robustness of classification, some techniques such as data augmentation (including rotation, translation, scaling, etc.) and dropout regularization techniques can also be used. These techniques help to improve the generalization ability of the model during training and reduce the overfitting of the model to the training data.

[0110] In some embodiments, classification is performed in combination with multi-modal data. Specifically, in addition to image features, other information related to images can also be introduced, such as sensor data, environmental information, etc. These multi-modal data are processed through joint learning, which can further improve the accuracy of classification.

[0111] After classification, a prediction result of the garbage class is finally output. This prediction result will be fed back to the system for subsequent actions, such as garbage classification, detection, or dispensing, etc.

[0112] The environmental image includes at least one dimension of data selected from an RGB image, a depth image, or multi-dimensional image data; the quantum optics enhanced visual lens system includes a quantum superposition state light source and a quantum entangled state light source, by adjusting the phase and amplitude of the light source, an enhanced light field distribution in the image acquisition process is formed;

[0113] Tensor decomposition is achieved through high-order tensor decomposition method, CANDECOMP / PARAFAC decomposition is adopted;

[0114] The CANDECOMP / PARAFAC decomposition method extracts the main spatial features and texture information in the image through step-by-step optimization, and stores them in the form of low-dimensional matrices, reducing data redundancy;

[0115] The steps of tensor decomposition include:

[0116] The image data is organized into a high-dimensional tensor, each dimension in the tensor representing a different attribute or spatial feature of the image;

[0117] The high-dimensional tensor is applied to CANDECOMP / PARAFAC decomposition, which decomposes the tensor into a series of low-dimensional matrices, each representing an important feature of the original data;

[0118] The low-dimensional matrices are further processed by a deep neural network to ensure that the garbage objects in the image are effectively identified and classified;

[0119] The path optimization algorithm is based on the variational method and the steepest descent method for path optimization, and the steps of path optimization include:

[0120] Based on the consideration of the environmental map and obstacles, a kinematic model of the robot is established, and the path is optimized by the variational method to calculate the best driving trajectory;

[0121] The steepest descent method is used to adjust the path in real time, and the path is corrected according to the changes in dynamic environmental data, so that the robot can efficiently reach the target position for garbage picking;

[0122] The environmental data in path optimization includes the current position of the robot, the position of obstacles in the environment, and the expected target position;

[0123] The path optimization updates the environmental data in real time, optimizes the path cost function by combining the variational method, and corrects the path by combining the steepest descent method, so that the robot can perform garbage picking tasks in unknown or dynamically changing environments;

[0124] The quantum annealing method is used for training the classification model;

[0125] The training process includes optimizing the objective function by the quantum annealing method;

[0126] The quantum annealing method adjusts the parameters of the quantum state to improve the classification performance of the garbage classification model;

[0127] The objective function in the quantum optimization algorithm includes the classification loss function and the regularization term;

[0128] The objective function aims to minimize the classification error, which adjusts the parameters through quantum optimization to optimize the accuracy of garbage classification;

[0129] The training process of the classification model is executed on a quantum computing platform;

[0130] The quantum computing platform is equipped with a quantum processing unit responsible for quantum optimization of model parameters;

[0131] The quantum optimization process is realized by a quantum annealing method, mainly adjusting the weight and bias of the model;

[0132] The collection of image data is through a quantum optical enhanced visual lens system, the system adjusts according to the quantum characteristics of the light source, enhances the tiny details in the collected image, and makes the garbage features in the image more obvious.

[0133] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An intelligent garbage detection method based on light source visual lens recognition, characterized in that, The method comprises the following steps: S1, capturing garbage and environment image data through a visual sensor; S2, collecting the image data using a quantum optics enhanced visual lens system; S3, performing tensor decomposition on the acquired image data, converting the multi-dimensional image data into a low-dimensional matrix using the CANDECOMP / PARAFAC decomposition method, and obtaining a compressed data representation; S4, inputting the converted low-dimensional matrix into a deep neural network to train a classification model, identifying garbage objects in the image first, and then performing feature extraction and classification; S5, calculating the robot path through a path optimization algorithm based on the environment image data, allowing the robot to complete the garbage grabbing task in a changing environment; The environment image includes at least one dimension of data selected from an RGB image, a depth image, or multi-dimensional image data; The quantum optics enhanced visual lens system includes a quantum superposition state light source and a quantum entangled state light source, which forms an enhanced light field distribution in the image acquisition process by adjusting the phase and amplitude of the light source; The path optimization algorithm optimizes the path based on the variational method and the steepest descent method, and the path optimization steps include: Based on the consideration of the environment map and obstacles, a kinematic model of the robot is established, the path is optimized through the variational method, and the best driving trajectory is calculated; The steepest descent method is used to adjust the path in real time, the path is corrected according to the dynamic environment data changes, and the robot is driven to efficiently reach the target position for garbage picking. 2.The intelligent garbage detection method based on light source visual lens recognition of claim 1, wherein, The tensor decomposition is realized by a high-order tensor decomposition method, and the CANDECOMP / PARAFAC decomposition is used; The CANDECOMP / PARAFAC decomposition method extracts the main spatial features and texture information in the image through step-by-step optimization and stores it in the form of a low-dimensional matrix, reducing data redundancy. 3.The intelligent garbage detection method based on light source visual lens recognition of claim 1, wherein, The steps of the tensor decomposition include: Organize the image data into a high-dimensional tensor, and each dimension in the tensor represents a different attribute or spatial feature of the image; Apply CANDECOMP / PARAFAC decomposition to the high-dimensional tensor to decompose the tensor into a series of low-dimensional matrices, each matrix representing an important feature of the original data; The low-dimensional matrix is further processed by a deep neural network after decomposition to ensure that garbage objects in the image are effectively identified and classified. 4.The intelligent garbage detection method based on light source visual lens recognition of claim 1, wherein, The environmental data in the path optimization includes the current position of the robot, the position of obstacles in the environment, and the expected target position; The path optimization updates the environmental data in real time, optimizes the path cost function using the variational method, and corrects the path using the steepest descent method, allowing the robot to perform garbage picking tasks in unknown or dynamically changing environments. 5.The intelligent garbage detection method based on light source visual lens recognition of claim 1, wherein, The training of the classification model uses a quantum annealing method; The training process includes optimizing the objective function using the quantum annealing method; The quantum annealing method improves the classification performance of the garbage classification model by adjusting the parameters of the quantum state. 6.The intelligent garbage detection method based on light source visual lens recognition of claim 5, characterized in that, The objective function in the quantum annealing method includes a classification loss function and a regularization term; The objective function aims to minimize the classification error, which adjusts the parameters through quantum optimization to optimize the accuracy of garbage classification. 7.The intelligent garbage detection method based on light source visual lens recognition of claim 1, wherein, The training process of the classification model is executed by a quantum computing platform; The quantum computing platform is configured with a quantum processing unit responsible for quantum optimization of model parameters; The quantum optimization process is realized by a quantum annealing method, which mainly adjusts the weights and biases of the model. 8.The intelligent garbage detection method based on light source visual lens recognition of claim 1, wherein, The image data acquisition is realized by a quantum optical enhanced visual lens system, which adjusts according to the quantum characteristics of the light source, enhances the micro details in the collected image, and makes the garbage features in the image more obvious.

Images