CN-121982353-A - Renewable resource sorting control method and system based on artificial intelligence

Abstract

The invention provides a renewable resource sorting control method and system based on artificial intelligence, wherein the method comprises the steps of extracting characteristics of an acquired renewable resource image to obtain an image characteristic vector; the method comprises the steps of classifying and identifying the renewable resources according to the image feature vectors to obtain category identification of the renewable resources, detecting targets of the renewable resources according to the image feature vectors to obtain boundary frame coordinates and grabbing point coordinates of the renewable resources, generating sorting action instructions of the manipulator according to the category identification, the boundary frame coordinates and the grabbing point coordinates, and carrying out resource sorting control on the manipulator according to the sorting action instructions. The automatic identification and the accurate sorting of the renewable resources are realized through the artificial intelligence technology, the sorting efficiency and the accuracy are improved, the labor cost is reduced, and the automatic sorting device is suitable for automatic sorting treatment of various renewable resources.

Inventors

• ZHAO XIAOFENG

Assignees

• 深圳能源环保股份有限公司

Dates

Publication Date

20260505

Application Date

20251201

Claims (8)

1. 1. The renewable resource sorting control method based on artificial intelligence is characterized by comprising the following steps of: performing feature extraction processing on the acquired renewable resource images to obtain image feature vectors; Classifying and identifying the renewable resources according to the image feature vector to obtain a category identification of the renewable resources; Performing target detection on the renewable resources according to the image feature vector to obtain boundary frame coordinates and grabbing point coordinates of the renewable resources; and generating a sorting action instruction of the manipulator according to the category identification, the boundary box coordinates and the grabbing point coordinates, and carrying out resource sorting control on the manipulator according to the sorting action instruction.
2. 2. The artificial intelligence based renewable resource sorting control method of claim 1, wherein the image feature vector comprises a local feature vector and a global feature vector; The feature extraction processing is performed on the collected renewable resource image to obtain an image feature vector, which comprises the following steps: and extracting features based on the renewable resource image through a lightweight hybrid neural network to obtain the local feature vector and the global feature vector.
3. 3. The artificial intelligence based renewable resource sorting control method according to claim 2, wherein the obtaining the local feature vector and the global feature vector by feature extraction based on the renewable resource image through a lightweight hybrid neural network comprises: carrying out local feature extraction on the renewable resource image through a convolutional neural network module in the lightweight hybrid neural network to obtain the local feature vector; Performing global feature extraction on the renewable resource image through a position-preserving visual transducer module in the lightweight hybrid neural network to obtain the global feature vector; And obtaining the image feature vector based on the local feature vector and the global feature vector.
4. 4. The artificial intelligence based renewable resource sorting control method according to claim 3, wherein the obtaining the local feature vector by performing local feature extraction on the renewable resource image through a convolutional neural network module in the lightweight hybrid neural network includes: Performing low-dimensional to high-dimensional channel expansion processing based on the renewable resource image through an expansion convolution layer in the convolution neural network module to obtain a high-dimensional expansion feature map; performing depth convolution and nonlinear transformation processing on the basis of the high-dimensional expansion feature map through a depth separable convolution layer and an activation function in the convolution neural network module to obtain an activation feature map; performing high-dimensional to low-dimensional channel compression processing based on the activation feature map through a linear compression convolutional layer in a convolutional neural network module to obtain a low-dimensional compression feature map; and when the input and output dimensions are matched, performing jump connection processing on the basis of the low-dimensional compression feature map and the renewable resource image to obtain the local feature vector.
5. 5. The artificial intelligence based renewable resource sorting control method according to claim 3, wherein the performing global feature extraction on the renewable resource image by the position preserving visual transformer module in the lightweight hybrid neural network, to obtain the global feature vector includes: performing unfolding operation on the renewable resource image through a position-keeping visual transducer module in the lightweight hybrid neural network to obtain a pixel sequence; performing correlation calculation based on the pixel sequence through a separated self-attention mechanism to obtain an attention characteristic sequence; performing intra-patch correlation calculation based on the pixel sequence through a patch attention mechanism to obtain a patch characteristic sequence; And carrying out fusion processing based on the attention feature sequence and the patch feature sequence, and recovering space position information through folding operation to obtain a global feature vector.
6. 6. The method for controlling sorting of renewable resources based on artificial intelligence according to claim 1, wherein the performing object detection on the renewable resources according to the image feature vector to obtain bounding box coordinates and grabbing point coordinates of the renewable resources comprises: Performing target detection head network processing on the image feature vector to obtain a candidate region feature map; Carrying out bounding box regression calculation processing on the candidate region feature map to obtain bounding box coordinates of the renewable resources; Performing key point detection processing on the candidate region feature map to obtain initial grabbing point coordinates of the renewable resources; And carrying out space correction processing on the initial grabbing point coordinates according to the boundary frame coordinates to obtain corrected grabbing point coordinates.
7. 7. The method for controlling sorting of renewable resources based on artificial intelligence according to claim 1, wherein the generating a sorting action instruction of a manipulator according to the category identification, the bounding box coordinates and the grabbing point coordinates and controlling the sorting of resources of the manipulator according to the sorting action instruction comprises: Inquiring the position of the discharge port according to the category identification and a preset category discharge port mapping table to obtain a target discharge port coordinate; performing obstacle detection and collision avoidance processing based on the boundary frame coordinates to obtain a safe motion space; Performing track planning processing based on the grabbing point coordinates, the target discharge port coordinates and the safe motion space to obtain a collision-free motion track; And generating an action sequence according to the collision-free motion track to obtain the sorting action instruction, and controlling the resource sorting of the manipulator according to the sorting action instruction.
8. 8. An artificial intelligence based renewable resource sorting control system, comprising: The characteristic extraction unit is used for carrying out characteristic extraction processing on the acquired renewable resource images to obtain image characteristic vectors; The classification and identification unit is used for classifying and identifying the renewable resources according to the image feature vectors to obtain category identifiers of the renewable resources; The target detection unit is used for carrying out target detection on the renewable resources according to the image feature vector to obtain the boundary frame coordinates and the grabbing point coordinates of the renewable resources; And the sorting control unit is used for generating a sorting action instruction of the manipulator according to the category identification, the boundary box coordinates and the grabbing point coordinates and carrying out resource sorting control on the manipulator according to the sorting action instruction.

Description

Renewable resource sorting control method and system based on artificial intelligence Technical Field The invention relates to the technical field of image recognition, in particular to a renewable resource sorting control method and system based on artificial intelligence. Background With the enhancement of environmental awareness and the increase of resource recovery demands, the renewable resource sorting technology becomes a key link of the development of circular economy. Traditional renewable resource sorting mainly relies on manual identification and classification, and has the problems of low efficiency, high cost, unstable sorting precision and the like. While some automatic sorting devices based on machine vision are presented in the prior art, these devices typically employ simple color recognition or shape matching algorithms for sorting materials. The existing sorting system has low accuracy in identifying renewable resources of different materials under a complex environment, and particularly for materials with similar appearance and different materials, such as different types of plastic products, can not be accurately distinguished, so that the sorting error rate is high, and the quality and efficiency of resource recovery are affected. Disclosure of Invention The first aspect of the invention provides an artificial intelligence-based renewable resource sorting control method, which comprises the following steps: performing feature extraction processing on the acquired renewable resource images to obtain image feature vectors; Classifying and identifying the renewable resources according to the image feature vector to obtain a category identification of the renewable resources; Performing target detection on the renewable resources according to the image feature vector to obtain boundary frame coordinates and grabbing point coordinates of the renewable resources; and generating a sorting action instruction of the manipulator according to the category identification, the boundary box coordinates and the grabbing point coordinates, and carrying out resource sorting control on the manipulator according to the sorting action instruction. Further, the image feature vector includes a local feature vector and a global feature vector; The feature extraction processing is performed on the collected renewable resource image to obtain an image feature vector, which comprises the following steps: and extracting features based on the renewable resource image through a lightweight hybrid neural network to obtain the local feature vector and the global feature vector. Further, the obtaining the local feature vector and the global feature vector by performing feature extraction based on the renewable resource image through a lightweight hybrid neural network includes: carrying out local feature extraction on the renewable resource image through a convolutional neural network module in the lightweight hybrid neural network to obtain the local feature vector; Performing global feature extraction on the renewable resource image through a position-preserving visual transducer module in the lightweight hybrid neural network to obtain the global feature vector; And obtaining the image feature vector based on the local feature vector and the global feature vector. Further, the performing, by the convolutional neural network module in the lightweight hybrid neural network, local feature extraction on the renewable resource image, to obtain the local feature vector includes: Performing low-dimensional to high-dimensional channel expansion processing based on the renewable resource image through an expansion convolution layer in the convolution neural network module to obtain a high-dimensional expansion feature map; performing depth convolution and nonlinear transformation processing on the basis of the high-dimensional expansion feature map through a depth separable convolution layer and an activation function in the convolution neural network module to obtain an activation feature map; performing high-dimensional to low-dimensional channel compression processing based on the activation feature map through a linear compression convolutional layer in a convolutional neural network module to obtain a low-dimensional compression feature map; and when the input and output dimensions are matched, performing jump connection processing on the basis of the low-dimensional compression feature map and the renewable resource image to obtain the local feature vector. Further, the performing global feature extraction on the renewable resource image by the position preserving visual transformer module in the lightweight hybrid neural network, to obtain the global feature vector includes: performing unfolding operation on the renewable resource image through a position-keeping visual transducer module in the lightweight hybrid neural network to obtain a pixel sequence; performing correlation calculation based on the pixel sequenc