A method for crushing and treating oily waste

Abstract

The present application relates to the field of intelligent control, and provides a method for crushing and processing oily waste, which uses data analysis and image processing technology based on deep learning to embed and encode the property parameter data such as oil content, viscosity and hardness of the dried oily waste, and extract multi-scale state features and perform significant joint perception fusion on the detection state image of the dried oily waste, so as to automatically obtain the recommended value of the crushing speed based on the cross-modal fusion characterization features between the embedded coding features of the property parameter data and the multi-scale fusion features of the detection state image. In this way, the crushing parameters can be adjusted in time according to the specific characteristics of the current material and the state changes of the waste, thereby improving the flexibility and intelligence of the crushing and processing of the oily waste.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to a method for crushing and processing oily waste. Background Art

[0002] With the acceleration of industrialization, especially in the petroleum and chemical industries, the amount of oily waste generated by underground operations continues to increase. These wastes mainly include PVC, plastic bags and domestic garbage. These wastes not only contain toxic and hazardous substances, but also may contain flammable and explosive oil substances, which pose a huge threat to the safety of underground operations and the environment. Once these wastes are improperly handled, they are very likely to cause fires or explosions when exposed to open flames or high temperatures, endangering the lives of underground workers and the safety of production equipment. Therefore, safe and reliable crushing and transportation technology is of great significance to eliminating safety hazards, ensuring the safety of underground operations and environmental protection.

[0003] Crushing plays a key role in the transportation process of oily waste. It reduces the volume and shape of the material to facilitate subsequent transportation and processing. This process not only improves the material processing efficiency, but also helps to release the oil substances attached to the waste that are difficult to directly recover, providing a basis for resource recovery and environmental protection.

[0004] However, the traditional crushing treatment method of oily waste mainly relies on empirical operations and fixed parameter settings, and cannot be dynamically adjusted according to the specific property parameters of the dried oily waste (such as oil content, viscosity and hardness), which means that when facing wastes with different characteristics, the crushing effect may not be satisfactory. In addition, due to the lack of real-time monitoring and analysis of property parameter data, the traditional method cannot adjust the crushing parameters in time according to the state changes of the waste, resulting in insufficient sensitivity and effectiveness in dealing with sudden changes or when immediate response is required, thereby reducing the flexibility and adaptability of the treatment.

[0005] Therefore, an optimized method for crushing and processing oily waste is desired. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present application provides a method for crushing and processing oily waste.

[0007] According to one aspect of the present application, a method for crushing and processing oily waste is provided, which comprises: preparing oily waste; drying the oily waste to obtain dried oily waste; conveying the dried oily waste to a crushing device for crushing to obtain crushed oily waste; wherein conveying the dried oily waste to a crushing device for crushing to obtain crushed oily waste comprises: obtaining property parameter data of the dried oily waste, wherein the property parameter data comprises oil content, viscosity and hardness; obtaining a detection state image of the dried oily waste; passing the detection state image of the dried oily waste through an oily waste state feature extractor to obtain a shallow state feature map of the oily waste and an oily waste state feature map; The method comprises the following steps: a first step is to generate a semantic state feature map of the oil-stained waste; a second step is to generate a semantic state feature map of the oil-stained waste; a third step is to generate a semantic state feature map of the oil-stained waste; a fourth step is to generate a semantic state feature map of the oil-stained waste; a fourth step is to generate a semantic state feature map of the oil-stained waste; a fifth step is to generate a semantic state feature map of the oil-stained waste; a fifth step is to generate a semantic state feature map of the oil-stained waste; a sixth step is to generate a semantic state feature map of the oil-stained waste; a sixth step is to generate a semantic state feature map of the oil-stained waste; a sixth step is to generate a semantic state feature map of the oil-stained waste; a sixth step is to generate a semantic state feature map of the oil-stained waste; a sixth step is to generate a semantic state feature map of the oil-stained waste; a seventh ...

[0008] In the above-mentioned method for crushing and processing oily waste, the detection state image of the dried oily waste is passed through an oily waste state feature extractor to obtain a shallow state feature map of the oily waste and a semantic state feature map of the oily waste, including: passing the detection state image of the dried oily waste through an oily waste state feature extractor based on a dilute pyramid network to obtain the shallow state feature map of the oily waste and the semantic state feature map of the oily waste.

[0009] In the above-mentioned oily waste crushing and processing method, the shallow state feature map of the oily waste and the semantic state feature map of the oily waste are passed through a salient feature joint perception module based on an attention mechanism to obtain a multi-scale joint perception fusion feature map of the oily waste state, including: reshaping the shallow state feature map of the oily waste and the semantic state feature map of the oily waste to obtain a shallow state feature shape reshaping matrix of the oily waste and a semantic state feature shape reshaping matrix of the oily waste; inputting the shallow state feature shape reshaping matrix of the oily waste and the semantic state feature shape reshaping matrix of the oily waste into a feature channel-by-channel interactive perception module to obtain a shallow-semantic dependency matrix of the oily waste and a semantic-shallow dependency matrix of the oily waste; inputting the shallow-semantic dependency matrix of the oily waste and the semantic-shallow dependency matrix of the oily waste into a random dropout module to obtain a pruned shallow-semantic dependency matrix of the oily waste. A semantic dependency matrix and a pruned oiled waste semantic-shallow dependency matrix; matrix multiplying the pruned oiled waste shallow-semantic dependency matrix with the oiled waste semantic state feature shape reshaping matrix to obtain a dependency-optimized oiled waste semantic state feature matrix; matrix multiplying the pruned oiled waste semantic-shallow dependency matrix with the oiled waste shallow state feature shape reshaping matrix to obtain a dependency-optimized oiled waste shallow state feature matrix; feature-reshaping the dependency-optimized oiled waste shallow state feature matrix and the dependency-optimized oiled waste semantic state feature matrix to obtain an optimized oiled waste shallow state feature map and an optimized oiled waste semantic state feature map; calculating the weighted sum of the optimized oiled waste shallow state feature map and the optimized oiled waste semantic state feature map to obtain the oiled waste state multi-scale joint perception fusion feature map.

[0010] In the above-mentioned crushing and processing method for oily waste, the shallow state feature shape reshaping matrix of the oily waste and the semantic state feature shape reshaping matrix of the oily waste are input into the feature channel-by-channel interactive perception module to obtain the shallow-semantic dependency matrix of the oily waste and the semantic-shallow dependency matrix of the oily waste, including: calculating the matrix product between the transposed matrix of the shallow state feature shape reshaping matrix of the oily waste and the shape reshaping matrix of the semantic state feature of the oily waste to obtain a first correlation feature matrix; calculating the positional division of the scale of the first correlation feature matrix and the first correlation feature matrix to obtain a first correlation feature modulation matrix; calculating the matrix product between the transposed matrix of the semantic state feature shape reshaping matrix of the oily waste and the shallow state feature shape reshaping matrix of the oily waste to obtain a second correlation feature matrix; calculating the positional division of the scale of the second correlation feature matrix and the second correlation feature matrix to obtain a second correlation feature modulation matrix; and inputting the first correlation feature modulation matrix and the second correlation feature modulation matrix into the Softmax function respectively to obtain the shallow-semantic dependency matrix of the oily waste and the semantic-shallow dependency matrix of the oily waste.

[0011] In the above-mentioned method for crushing and processing oily waste, the weighted sum of the optimized shallow state feature map of oily waste and the optimized semantic state feature map of oily waste is calculated to obtain the multi-scale joint perception fusion feature map of the oily waste state, including: multiplying the optimized shallow state feature map of oily waste and the optimized semantic state feature map of oily waste by the corresponding weight hyperparameters by position, and then adding the obtained optimized weighted shallow state feature map of oily waste and the optimized weighted semantic state feature map of oily waste by position to obtain the multi-scale joint perception fusion feature map of the oily waste state.

[0012] In the above-mentioned method for crushing and processing oily waste, the property parameter data of the dried oily waste is embedded and encoded to obtain the oily waste property embedded coding vector, including: using the property parameter embedded coding matrix to embed the property parameter data of the dried oily waste to obtain the oily waste property embedded coding vector.

[0013] In the above-mentioned method for crushing and treating oily waste, the oily waste property embedding coding vector and the multi-scale joint perception fusion feature map of the oily waste state are subjected to a Transformer-like structure encoder based on cross-modal feature fusion to obtain a multi-modal property fusion representation feature map of the oily waste, including: performing layer normalization and convolution coding on the multi-scale joint perception fusion feature map of the oily waste state to obtain a multi-scale joint perception space deepening feature map of the oily waste state; copying the multi-scale joint perception space deepening feature map of the oily waste state to obtain a backup multi-scale joint perception space deepening feature map of the oily waste state; and reshaping the backup multi-scale joint perception space deepening feature map of the oily waste state and the multi-scale joint perception space deepening feature map of the oily waste state respectively to obtain a backup multi-scale joint perception space deepening feature matrix of the oily waste state and a multi-scale joint perception space deepening feature matrix of the oily waste state. The spatial deepening feature matrix is ​​used as the imitation query feature matrix and the imitation key feature matrix; the product between the imitation query feature matrix and the transposed matrix of the imitation key feature matrix is ​​calculated to obtain the multi-scale joint perception local autocorrelation library matrix of the oiled waste state; the multi-scale joint perception local autocorrelation library matrix of the oiled waste state is input into the attention gating unit based on the GELU function to obtain the multi-scale joint perception gated local autocorrelation library weight matrix of the oiled waste state; the oiled waste property embedding coding vector is used as the value feature vector, and the product between the multi-scale joint perception gated local autocorrelation library weight matrix of the oiled waste state and the oiled waste property embedding coding vector is calculated to obtain the oiled waste property conditional weight vector; the oiled waste property conditional weight vector is used as the weight vector, and the multi-scale joint perception fusion feature map of the oiled waste state is weighted channel by channel to obtain the multi-modal property fusion representation feature map of the oiled waste.

[0014] In the above-mentioned method for crushing and processing oily waste, the multi-scale joint perception fusion feature map of the oily waste state is subjected to layer normalization and convolution coding processing to obtain a multi-scale joint perception spatial deepening feature map of the oily waste state, including: performing layer normalization processing on the multi-scale joint perception fusion feature map of the oily waste state to obtain a layer-normalized multi-scale joint perception fusion feature map of the oily waste state; performing point convolution coding and spatial convolution coding on the layer-normalized multi-scale joint perception fusion feature map of the oily waste state to obtain the multi-scale joint perception spatial deepening feature map of the oily waste state.

[0015] In the above-mentioned method for crushing and processing oily waste, a control result is obtained based on the characteristic map of multimodal properties fusion characterization of the oily waste, including: passing the characteristic map of multimodal properties fusion characterization of the oily waste through a crusher speed controller based on a decoder to obtain the control result.

[0016] The present application has significant technical effects due to the adoption of the above technical solutions: the crushing and processing method of oily waste provided by the present application adopts data analysis and image processing technology based on deep learning to embed and encode the property parameter data such as oil content, viscosity and hardness of the dried oily waste, and extracts multi-scale state features and performs significant joint perception fusion on the detection state image of the dried oily waste, thereby automatically obtaining the recommended value of the crushing speed based on the cross-modal fusion characterization features between the embedded coding features of the property parameter data and the multi-scale fusion features of the detection state image. In this way, the crushing parameters can be adjusted in time according to the specific characteristics of the current material and the state changes of the waste, thereby improving the flexibility and intelligence of the crushing and processing of the oily waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 Flow chart of a method for crushing and treating oily waste according to an embodiment of the present application.

[0019] Figure 2 The present invention is a flow chart of a method for crushing oily waste according to an embodiment of the present application, wherein the dried oily waste is transported to a crushing device for crushing to obtain crushed oily waste.

[0020] Figure 3 This is a data flow diagram of the oily waste after being dried, which is transported to a crushing device for crushing to obtain the crushed oily waste in the method for crushing oily waste according to an embodiment of the present application.

[0021] Figure 4 In the method for crushing and processing oily waste according to an embodiment of the present application, the shallow state feature map of the oily waste and the semantic state feature map of the oily waste are subjected to a salient feature joint perception module based on an attention mechanism to obtain a flow chart of a multi-scale joint perception fusion feature map of the oily waste state.

[0022] Figure 5In the method for crushing and processing oily waste according to an embodiment of the present application, the oily waste property is embedded in the coding vector and the multi-scale joint perception fusion feature map of the oily waste state is passed through a Transformer-like structure encoder based on cross-modal feature fusion to obtain a flowchart of the multi-modal property fusion representation feature map of the oily waste. DETAILED DESCRIPTION

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0024] With the rapid development of industrialization, the amount of oily waste generated by industries such as petrochemicals has increased dramatically. These wastes include PVC, plastic bags, and domestic garbage, which contain harmful substances and flammable and explosive oil substances, which pose a serious threat to underground operation safety and the environment. Therefore, the development of safe and efficient crushing and transportation technology is crucial to reduce safety hazards and protect the safety of operators and equipment.

[0025] Crushing technology plays a core role in the treatment of oily waste. It reduces the volume and shape of materials, improves processing efficiency, and helps to recover attached oil substances, laying the foundation for resource recycling and environmental protection. However, traditional crushing methods rely on empirical operations and fixed parameters, lack the ability to dynamically adjust the characteristics of waste, resulting in unstable treatment effects. In addition, due to the lack of real-time monitoring and analysis of oily waste, traditional methods are not responsive enough when dealing with emergencies or situations that require timely response, affecting the flexibility and adaptability of treatment.

[0026] Based on this, the present application provides a method for crushing and processing oily waste. Figure 1 FIG. 1 is a flow chart of a method for crushing and treating oily waste according to an embodiment of the present application. Figure 1 As shown, the method for crushing and processing oily waste according to an embodiment of the present application includes: S110, preparing oily waste; S120, drying the oily waste to obtain dried oily waste; S130, transporting the dried oily waste to a crushing device for crushing to obtain crushed oily waste.

[0027] Accordingly, in the process of conveying the dried oily waste to the crushing equipment for crushing to obtain the crushed oily waste, the technical concept of the present application is to obtain the property parameter data (oil content, viscosity and hardness) of the dried oily waste, and obtain the detection state image of the dried oily waste, and use data analysis and image processing technology based on deep learning to embed the property parameter data, extract multi-scale state features and perform significant joint perception fusion on the detection state image, so as to automatically obtain the recommended value of the crushing speed based on the cross-modal fusion characterization features between the embedded coding features of the property parameter data and the multi-scale fusion features of the detection state image. In this way, the crushing parameters can be adjusted in time according to the specific characteristics of the current material (such as oil content, viscosity and hardness) and the state change of the waste, so as to ensure that the crushing process is more efficient and reliable, avoid the problem of incomplete or excessive crushing, and significantly improve the flexibility and intelligence of the crushing treatment of oily waste.

[0028] Figure 2 The present invention is a flow chart of a method for crushing oily waste according to an embodiment of the present application, wherein the dried oily waste is transported to a crushing device for crushing to obtain crushed oily waste. Figure 3 The data flow diagram of the oily waste after being dried is shown in the crushing method according to the embodiment of the present application, wherein the oily waste after being dried is transported to the crushing equipment for crushing to obtain the crushed oily waste. Figure 2 and Figure 3As shown, the dried oily waste is transported to a crushing device for crushing to obtain crushed oily waste, including: S131, obtaining property parameter data of the dried oily waste, wherein the property parameter data includes oil content, viscosity and hardness; S132, obtaining a detection state image of the dried oily waste; S133, passing the detection state image of the dried oily waste through an oily waste state feature extractor to obtain a shallow state feature map of the oily waste and a semantic state feature map of the oily waste; S134, passing the shallow state feature map of the oily waste and the semantic state feature map of the oily waste through an attention-based feature extraction method. A significant feature joint perception module is used to obtain a multi-scale joint perception fusion feature map of the oily waste state; S135, embedding and encoding the property parameter data of the dried oily waste to obtain an embedded coding vector of the oily waste property; S136, passing the oily waste property embedded coding vector and the multi-scale joint perception fusion feature map of the oily waste state through a Transformer-like structure encoder based on cross-modal feature fusion to obtain a multi-modal property fusion representation feature map of the oily waste; S137, based on the multi-modal property fusion representation feature map of the oily waste, a control result is obtained, and the control result is used to represent the recommended value of the crushing rotation speed.

[0029] In step S131 and step S132, the property parameter data of the dried oily waste is obtained, wherein the property parameter data includes oil content, viscosity and hardness, and the detection state image of the dried oily waste is obtained. It should be understood that the oil content refers to the proportion of oil substances in the waste. Waste with high oil content may release more oil when crushed, increasing the risk of fire. The crushing speed should be appropriately reduced to reduce the release of oil and avoid safety hazards. On the contrary, waste with low oil content can use a higher crushing speed to improve the processing efficiency. Viscosity refers to the fluidity of oil substances in the waste. High-viscosity waste may require a lower crushing speed to ensure that the crushing equipment can effectively handle these materials and avoid poor material flow or equipment failure due to too fast a speed. Low-viscosity waste can use a higher speed to improve the processing efficiency. Hardness refers to the compressive and wear resistance of the waste material. For waste with high hardness, the crushing speed needs to be appropriately increased to provide sufficient energy for effective crushing. The detection state image of the oily waste after drying can provide information such as the shape, size, and surface state of the waste, which can reflect the physical state of the waste and help identify the aggregation or dispersion of the waste. When the detection state image shows that the waste is aggregated, it may be necessary to reduce the crushing speed to avoid equipment clogging; and when the detection state image shows that the waste is well dispersed, the speed can be appropriately increased to enhance the processing efficiency. In general, the property parameter data and the detection state image of the oily waste after drying provide important information, which can help dynamically adjust the crushing speed, thereby improving the processing efficiency and ensuring safety. In particular, in a specific embodiment of the present application, a sample can be selected from the processing process of the oily waste, and then the Soxhlet extraction method, the rotational viscometer and the Shore hardness tester are used to obtain the oil content, viscosity and hardness data of the oily waste after drying, and the detection state image of the oily waste after drying can be obtained by a high-resolution camera.

[0030] In step S133, the detection state image of the dried oily waste is passed through an oily waste state feature extractor to obtain a shallow state feature map of the oily waste and a semantic state feature map of the oily waste. Specifically, in an embodiment of the present application, the detection state image of the dried oily waste is passed through an oily waste state feature extractor to obtain a shallow state feature map of the oily waste and a semantic state feature map of the oily waste, including: passing the detection state image of the dried oily waste through an oily waste state feature extractor based on a dilute pyramid network to obtain the shallow state feature map of the oily waste and the semantic state feature map of the oily waste. It should be understood that the detection state image of the oily waste contains feature information of different scales, such as cracks, contours and surface textures of shallow oily waste and more abstract high-level semantic information. The dilated pyramid network can not only capture a wider range of information in the input image through the dilated convolution kernel, but also capture feature information of different scales at the same time, which is conducive to a more comprehensive and accurate understanding of the state of the oily waste, and can provide a scientific basis for subsequent crushing and processing to improve the efficiency and safety of crushing and processing. Based on this, in the technical solution of the present application, the detection state image of the dried oily waste is passed through an oily waste state feature extractor based on a dilated pyramid network to capture and extract features of different scales and levels, and obtain a shallow state feature map of the oily waste and a semantic state feature map of the oily waste.

[0031] In step S134, the shallow state feature map of oiled waste and the semantic state feature map of oiled waste are passed through a salient feature joint perception module based on an attention mechanism to obtain a multi-scale joint perception fusion feature map of the state of oiled waste. Accordingly, considering that the shallow state feature map of oiled waste and the semantic state feature map of oiled waste respectively express the features of different scales of oiled waste, and in order to simultaneously focus on the features of different scales, so as to capture the local details and global context information of oiled waste, thereby improving the expression ability of the overall features, in the technical solution of the present application, the shallow state feature map of oiled waste and the semantic state feature map of oiled waste are passed through a salient feature joint perception module based on an attention mechanism to obtain a multi-scale joint perception fusion feature map of the state of oiled waste. It is worth noting that the salient feature joint perception module based on the attention mechanism adopts a dual interactive attention mechanism to enhance the features of each scale, thereby improving the expression ability of the overall features. In particular, through this dual attention interaction, the module can more effectively highlight important features and suppress unimportant information, thereby obtaining richer and more representative information in the feature fusion process.

[0032] Specifically, the shallow state feature map of the oiled waste and the semantic state feature map of the oiled waste are first reshaped to retain the channel and spatial features of the original features, and the shallow state feature shape reshaping matrix of the oiled waste and the semantic state feature shape reshaping matrix of the oiled waste are obtained, which provide a basis for the subsequent interaction between the two. Afterwards, the dependency between the two reshaped feature matrices is calculated respectively by the feature channel-by-channel interaction perception module to reveal the intrinsic connection and interaction between the different features of the oiled waste state, and the shallow-semantic dependency matrix of the oiled waste and the semantic-shallow dependency matrix of the oiled waste are obtained. Then, the two dependency matrices are input into the random dropout module for pruning operation to identify and remove those feature channels that do not contribute much to the model performance or are completely redundant, reduce the complexity of the model, and avoid performance degradation caused by feature redundancy. It should be understood that the random dropout technology enables the model to maintain stability in the face of data perturbations by randomly ignoring or suppressing certain feature interactions. By removing unnecessary features through pruning, the overfitting phenomenon of the model on the training data can be reduced, because overfitting is often due to the model learning noise and outliers in the data. Then, the pruned features are matrix-multiplied with the reshaped features to strengthen the role of important or more discriminative features in the model, and the dependency-optimized shallow state feature matrix of oiled waste and the dependency-optimized semantic state feature matrix of oiled waste are obtained. Finally, the optimized features are reshaped to restore them to the form of feature maps, and weighted fusion is performed to obtain the final multi-scale joint perception fusion feature map of oiled waste state, which contains original features of different scales and the interactive features between the two.

[0033] Specifically, Figure 4 The present invention is a flowchart of a method for crushing and treating oily waste according to an embodiment of the present application, wherein the shallow state feature map of the oily waste and the semantic state feature map of the oily waste are subjected to a salient feature joint perception module based on an attention mechanism to obtain a multi-scale joint perception fusion feature map of the oily waste state. Figure 4As shown, the shallow state feature map of the oiled waste and the semantic state feature map of the oiled waste are passed through a salient feature joint perception module based on an attention mechanism to obtain a multi-scale joint perception fusion feature map of the oiled waste state, including: S1341, reshaping the shallow state feature map of the oiled waste and the semantic state feature map of the oiled waste to obtain a shallow state feature shape reshaping matrix of the oiled waste and a semantic state feature shape reshaping matrix of the oiled waste; S1342, inputting the shallow state feature shape reshaping matrix of the oiled waste and the semantic state feature shape reshaping matrix of the oiled waste into a feature channel-by-channel interactive perception module to obtain an oiled waste shallow-semantic dependency matrix and an oiled waste semantic-shallow dependency matrix; S1343, inputting the shallow-semantic dependency matrix of the oiled waste and the semantic-shallow dependency matrix of the oiled waste into a random dropout module to obtain a pruned shallow-semantic dependency matrix of the oiled waste and a pruned shallow-semantic dependency matrix of the oiled waste. S1344, matrix multiplying the pruned oiled waste shallow-semantic dependency matrix with the oiled waste semantic state feature shape reshaping matrix to obtain a dependency-optimized oiled waste semantic state feature matrix; S1345, matrix multiplying the pruned oiled waste semantic-shallow dependency matrix with the oiled waste shallow state feature shape reshaping matrix to obtain a dependency-optimized oiled waste shallow state feature matrix; S1346, reshaping the dependency-optimized oiled waste shallow state feature matrix and the dependency-optimized oiled waste semantic state feature matrix to obtain an optimized oiled waste shallow state feature map and an optimized oiled waste semantic state feature map; S1347, calculating the weighted sum of the optimized oiled waste shallow state feature map and the optimized oiled waste semantic state feature map to obtain the oiled waste state multi-scale joint perception fusion feature map.

[0034] More specifically, in an embodiment of the present application, the shape reshaping matrix of the shallow state characteristics of the oiled waste and the shape reshaping matrix of the semantic state characteristics of the oiled waste are input into the feature channel-by-channel interactive perception module to obtain the shallow-semantic dependency matrix of the oiled waste and the semantic-shallow dependency matrix of the oiled waste, including: calculating the matrix product between the transposed matrix of the shallow state characteristics of the oiled waste shape reshaping matrix and the shape reshaping matrix of the semantic state characteristics of the oiled waste to obtain a first correlation feature matrix; calculating the positional division of the scales of the first correlation feature matrix and the first correlation feature matrix to obtain a first correlation feature modulation matrix; calculating the matrix product between the transposed matrix of the semantic state characteristics of the oiled waste shape reshaping matrix and the shape reshaping matrix of the shallow state characteristics of the oiled waste to obtain a second correlation feature matrix; calculating the positional division of the scales of the second correlation feature matrix and the second correlation feature matrix to obtain a second correlation feature modulation matrix; and inputting the first correlation feature modulation matrix and the second correlation feature modulation matrix into the Softmax function respectively to obtain the shallow-semantic dependency matrix of the oiled waste and the semantic-shallow dependency matrix of the oiled waste.

[0035] More specifically, in an embodiment of the present application, the weighted sum of the optimized shallow state feature map of oiled waste and the optimized semantic state feature map of oiled waste is calculated to obtain the multi-scale joint perception fusion feature map of the oiled waste state, including: multiplying the optimized shallow state feature map of oiled waste and the optimized semantic state feature map of oiled waste by the corresponding weight hyperparameters by position, and then adding the obtained optimized weighted shallow state feature map of oiled waste and the optimized weighted semantic state feature map of oiled waste by position to obtain the multi-scale joint perception fusion feature map of the oiled waste state.

[0036] In an embodiment of the present application, specifically, the shallow state feature map of the oil-stained waste and the semantic state feature map of the oil-stained waste are passed through a salient feature joint perception module based on an attention mechanism to obtain a multi-scale joint perception fusion feature map of the oil-stained waste state, including: passing the shallow state feature map of the oil-stained waste and the semantic state feature map of the oil-stained waste through a salient feature joint perception module based on an attention mechanism, and processing them with the following salient feature joint perception formula to obtain the multi-scale joint perception fusion feature map of the oil-stained waste state; wherein the salient feature joint perception formula is: ;in, and are respectively the shallow state feature map of the oily waste and the semantic state feature map of the oily waste, For the reshape operation, and are respectively the shallow state feature shape reshaping matrix of the oily waste and the semantic state feature shape reshaping matrix of the oily waste, is the matrix product, and is the transposed matrix of the shallow state feature shape reshaping matrix of the oily waste and the semantic state feature shape reshaping matrix of the oily waste, and are the scales of the first correlation feature matrix and the second correlation feature matrix, respectively, for function, and are the oily waste shallow-semantic dependency matrix and the oily waste semantic-shallow dependency matrix, For pruning, and are the pruned oily waste shallow-semantic dependency matrix and the pruned oily waste semantic-shallow dependency matrix, respectively. and The dependency relationship is used to optimize the semantic state feature matrix of oily waste and the dependency relationship is used to optimize the shallow state feature matrix of oily waste, respectively. and are respectively the optimized shallow state feature map of oily waste and the optimized semantic state feature map of oily waste, and is the weight hyperparameter, It is a multi-scale joint perception fusion feature map of the oily waste state.

[0037] In step S135, embedding coding is performed on the property parameter data of the dried oily waste to obtain an embedded coding vector of the property of the oily waste. Specifically, in an embodiment of the present application, embedding coding is performed on the property parameter data of the dried oily waste to obtain an embedded coding vector of the property of the oily waste, including: embedding coding is performed on the property parameter data of the dried oily waste using a property parameter embedding coding matrix to obtain an embedded coding vector of the property of the oily waste. It should be understood that the property parameter data of the oily waste is usually of high dimension and also contains important characteristic information about the property. Based on this, in the technical solution of the present application, embedding coding is performed on the property parameter data of the dried oily waste using a property parameter embedding coding matrix to obtain an embedded coding vector of the property of the oily waste. In this way, embedding coding is performed by embedding coding through the property parameter embedding coding matrix, so that the property parameter data of the oily waste can be mapped to a lower dimensional vector space. This dimensionality reduction not only reduces the complexity of subsequent calculations, but also retains important characteristic information. And the property parameter embedding coding matrix can also capture the potential correlation between different property parameters. For example, there may be a relationship between oil content and viscosity, which can be effectively expressed through embedded coding, thereby achieving a deep understanding of the properties of oiled waste.

[0038] In step S136, the oil-stained waste property embedding coding vector and the oil-stained waste state multi-scale joint perception fusion feature map are passed through a Transformer-like structure encoder based on cross-modal feature fusion to obtain a multi-modal property fusion representation feature map of oil-stained waste. Accordingly, considering that the oil-stained waste property embedding coding vector expresses the intrinsic physical and chemical properties of oil-stained waste, the multi-scale joint perception fusion feature map of oil-stained waste state expresses the visual features of oil-stained waste at different scales, such as texture, shape and structure. In order to effectively realize the deep fusion between different modal data, so as to capture and utilize the unique information of different data sources, reveal the inherent connection and mutual influence between data, and thus improve the model's understanding and processing ability of oil-stained waste state, in the technical solution of the present application, the oil-stained waste property embedding coding vector and the oil-stained waste state multi-scale joint perception fusion feature map are passed through a Transformer-like structure encoder based on cross-modal feature fusion to obtain a multi-modal property fusion representation feature map of oil-stained waste. That is, the Transformer-like structure encoder based on cross-modal feature fusion draws on the self-attention mechanism in the Transformer architecture, that is, the input features are weighted through the self-attention mechanism, and the gating unit is used to control the information flow, so as to ensure that the features from different modalities can complement and enhance each other, thereby improving the quality of feature expression and the performance of the model. In detail, the multi-scale joint perception fusion feature map of the state of oil-stained waste is firstly subjected to layer normalization processing to ensure that different feature maps are consistent in scale, and the data deviation caused by the inconsistent scale is eliminated to obtain the layer-normalized multi-scale joint perception fusion feature map of the state of oil-stained waste. After that, the layer-normalized features are subjected to point convolution encoding and spatial convolution encoding to realize the integration of features and the extraction of spatial information to obtain the multi-scale joint perception spatial deepening feature map of the state of oil-stained waste, and the spatial features are retained to obtain the backup multi-scale joint perception spatial deepening feature map of the state of oil-stained waste. Next, the copied features and spatial features are reshaped to obtain the backup oil-stained waste state multi-scale joint perception spatial deepening feature matrix and the oil-stained waste state multi-scale joint perception spatial deepening feature matrix as the imitation query feature matrix and the imitation key feature matrix, and the autocorrelation matrix between the two is calculated to capture the oil-stained waste correlation between the two. Then, the attention gating unit based on the GELU function is used to adaptively adjust the autocorrelation weight of each eigenvalue in the autocorrelation matrix to achieve dynamic selection and enhancement of features to obtain the autocorrelation library weight matrix. Then, the oil-stained waste property embedding encoding vector and the local autocorrelation library weight matrix are matrix multiplied to realize the correlation relationship between cross-modalities to obtain the fusion condition weight vector.Finally, the fusion condition weight vector is used as the weight vector to weight the multi-scale joint perception fusion feature map of the oily waste state channel by channel to comprehensively characterize the multimodal information of the properties and state of the oily waste, and obtain the multimodal property fusion characterization feature map of the oily waste.

[0039] Specifically, Figure 5 In the method for crushing and treating oily waste according to an embodiment of the present application, the oily waste property embedding coding vector and the multi-scale joint perception fusion feature map of the oily waste state are passed through a Transformer-like structure encoder based on cross-modal feature fusion to obtain a flow chart of the multi-modal property fusion representation feature map of the oily waste. Figure 5 As shown, the oil-stained waste property embedding coding vector and the oil-stained waste state multi-scale joint perception fusion feature map are passed through a Transformer-like structure encoder based on cross-modal feature fusion to obtain an oil-stained waste multi-modal property fusion representation feature map, including: S1361, performing layer normalization and convolution coding processing on the oil-stained waste state multi-scale joint perception fusion feature map to obtain an oil-stained waste state multi-scale joint perception space deepening feature map; S1362, copying the oil-stained waste state multi-scale joint perception space deepening feature map to obtain a backup oil-stained waste state multi-scale joint perception space deepening feature map; S1363, respectively reshaping the backup oil-stained waste state multi-scale joint perception space deepening feature map and the oil-stained waste state multi-scale joint perception space deepening feature map to obtain a backup oil-stained waste state multi-scale joint perception space deepening feature matrix and an oil-stained waste state multi-scale joint perception space deepening feature matrix. is a simulated query feature matrix and a simulated key feature matrix; S1364, calculating the product between the simulated query feature matrix and the transposed matrix of the simulated key feature matrix to obtain a multi-scale joint perception local autocorrelation library matrix of the oiled waste state; S1365, inputting the multi-scale joint perception local autocorrelation library matrix of the oiled waste state into an attention gating unit based on the GELU function to obtain a multi-scale joint perception gated local autocorrelation library weight matrix of the oiled waste state; S1366, taking the oiled waste property embedding coding vector as the value feature vector, calculating the product between the multi-scale joint perception gated local autocorrelation library weight matrix of the oiled waste state and the oiled waste property embedding coding vector to obtain an oiled waste property conditional weight vector; S1367, taking the oiled waste property conditional weight vector as the weight vector, weighting the multi-scale joint perception fusion feature map of the oiled waste state channel by channel to obtain the multi-modal property fusion representation feature map of the oiled waste.

[0040] More specifically, in an embodiment of the present application, the multi-scale joint perception fusion feature map of the state of oiled waste is subjected to layer normalization and convolution coding processing to obtain a multi-scale joint perception spatial deepening feature map of the state of oiled waste, including: performing layer normalization processing on the multi-scale joint perception fusion feature map of the state of oiled waste to obtain a layer normalized multi-scale joint perception fusion feature map of the state of oiled waste; performing point convolution coding and spatial convolution coding on the layer normalized multi-scale joint perception fusion feature map of the state of oiled waste to obtain the multi-scale joint perception spatial deepening feature map of the state of oiled waste.

[0041] In an embodiment of the present application, specifically, the oily waste property embedding coding vector and the oily waste state multi-scale joint perception fusion feature map are passed through a Transformer-like structure encoder based on cross-modal feature fusion to obtain the oily waste multi-modal property fusion representation feature map, including: the oily waste property embedding coding vector and the oily waste state multi-scale joint perception fusion feature map are passed through a Transformer-like structure encoder based on cross-modal feature fusion, and processed with the following cross-modal formula to obtain the oily waste multi-modal property fusion representation feature map; wherein the cross-modal formula is: ;in, is the multi-scale joint perception fusion feature map of the oily waste state, is the layer normalization operation, is point convolutional coding, The convolution kernel is The convolution operation, The multi-scale joint perception space deepening feature map of the oily waste state is obtained. For copy operation, The multi-scale joint perception space deepening feature map of the backup oily waste state is provided. For the reshape operation, is the transpose of the feature matrix, is matrix multiplication, is the GELU function, is the weight matrix of the multi-scale joint perception gated local autocorrelation library of the oily waste state, is the oily waste property embedding coding vector, is the weight vector of the oily waste property conditions, It is a characteristic diagram representing the fusion of multimodal properties of the oily waste.

[0042] In step S137, a control result is obtained based on the multimodal property fusion characterization feature map of the oily waste, and the control result is used to represent the recommended value of the crushing speed. Specifically, in the embodiment of the present application, the control result is obtained based on the multimodal property fusion characterization feature map of the oily waste, including: passing the multimodal property fusion characterization feature map of the oily waste through a crusher speed controller based on a decoder to obtain the control result. That is, the multimodal property fusion characterization feature map of the oily waste obtained by cross-modal fusion of the oily waste property embedding coding vector and the multi-scale joint perception fusion feature map of the oily waste state is decoded to automatically obtain the recommended value of the crushing speed. In this way, the crushing parameters can be adjusted in time according to the specific characteristics of the current material (such as oil content, viscosity and hardness) and the state change of the waste, so as to ensure that the crushing process is more efficient and reliable, avoid the problem of incomplete crushing or excessive crushing, and significantly improve the flexibility and intelligence of the crushing treatment of the oily waste.

[0043] In particular, the present application takes into account that the embedded coding vector of the property of the oily waste and the multi-scale joint perception fusion feature map of the state of the oily waste respectively represent the embedded coding features of the property parameter data of the oily waste and the cross-scale and cross-depth saliency joint perception image semantic features of the detection state image of the oily waste. When they are passed through a Transformer-like structure encoder based on cross-modal feature fusion, the different dimensional association structures based on the cross-modal feature distribution will also have cross-modal fusion context association encoding differences, so that the multi-modal property fusion representation feature map of the oily waste has a complex spatial structure based on cross-modal joint coding, thereby expecting to improve its decoding regression convergence and generalization effects under complex spatial structures.

[0044] Therefore, in a preferred example of the present application, after passing the multimodal property fusion characterization feature map of oily waste through a crusher speed controller based on a decoder to obtain a control result, the multimodal property fusion characterization feature map of oily waste is optimized, specifically including: calculating the sum of the absolute values ​​of each eigenvalue of the multimodal property fusion characterization feature map of oily waste to obtain a first multimodal property fusion characterization spatial structure value of oily waste, and calculating the square root of the sum of the squares of each eigenvalue of the multimodal property fusion characterization feature map of oily waste to obtain a second multimodal property fusion characterization spatial structure value of oily waste, that is, ;in, The first characterization image representing the multimodal property fusion characterization feature map of the oily waste eigenvalues, represents a fusion characterization feature map of the multimodal properties of the oily waste, represents the set of real numbers, , and are the feature matrix width, feature matrix height and number of channels of the multimodal property fusion characterization feature map of the oily waste, respectively. represents the scale of the multimodal property fusion characterization feature map of the oily waste, It represents the spatial structure value of the multimodal property fusion representation of the first oily waste, It represents the spatial structure value of the fusion representation of the multimodal properties of the second oily waste.

[0045] Each eigenvalue of the oily waste multimodal property fusion characterization feature graph is multiplied by the first oily waste multimodal property fusion characterization spatial structure value and the second oily waste multimodal property fusion characterization spatial structure value respectively to obtain the first oily waste multimodal property fusion characterization structure reference value and the second oily waste multimodal property fusion characterization structure reference value corresponding to each eigenvalue.

[0046] Each eigenvalue of the oily waste multimodal property fusion characterization feature graph is multiplied by the scale of the oily waste multimodal property fusion characterization feature graph and the square root of the scale to obtain a first oily waste multimodal property fusion characterization scale transformation value and a second oily waste multimodal property fusion characterization scale transformation value corresponding to each eigenvalue.

[0047] The first oily waste multimodal property fusion representation structure reference value is divided by the difference between the first oily waste multimodal property fusion representation spatial structure value and the first oily waste multimodal property fusion representation scale transformation value to obtain the first oily waste multimodal property fusion representation transformation adjustment value.

[0048] The second oily waste multimodal property fusion representation structure reference value is divided by the difference between the second oily waste multimodal property fusion representation spatial structure value and the second oily waste multimodal property fusion representation scale transformation value to obtain the second oily waste multimodal property fusion representation transformation adjustment value.

[0049] The weighted sum of the first oily waste multimodal property fusion characterization transformation adjustment value and the second oily waste multimodal property fusion characterization transformation adjustment value is calculated to obtain each characteristic value of the optimized oily waste multimodal property fusion characterization feature map. Finally, the optimized oily waste multimodal property fusion characterization feature map is passed through a crusher speed controller based on a decoder to obtain a control result.

[0050] It is expressed as: ;in, The first characterization image representing the multimodal property fusion characterization feature map of the oily waste eigenvalues, , and are the feature matrix width, feature matrix height and number of channels of the multimodal property fusion characterization feature map of the oily waste, respectively. represents the scale of the multimodal property fusion characterization feature map of the oily waste, represents the set of real numbers, It represents the spatial structure value of the multimodal property fusion representation of the first oily waste, It represents the spatial structure value of the fusion representation of the multimodal properties of the second oily waste. represents the second oily waste multimodal property fusion characterization transformation adjustment value, represents the first oily waste multimodal property fusion characterization transformation adjustment value, represents a second oily waste multimodal property fusion representation transformation adjustment feature map composed of the second oily waste multimodal property fusion representation transformation adjustment values, represents a first oily waste multimodal property fusion representation transformation adjustment feature map composed of the first oily waste multimodal property fusion representation transformation adjustment values, represents the weighted hyperparameter, It means adding by position. It means point multiplication by position. Represents the optimized fusion characterization feature map of multimodal properties of oily waste.

[0051] That is, in this preferred example, for the spatial structure information of the feature set of the multimodal property fusion characterization feature map of the oily waste in the high-dimensional space, the scale-based frame transformation of each feature value of the multimodal property fusion characterization feature map of the oily waste is performed by taking the class norm space structured representation of the multimodal property fusion characterization feature map of the oily waste as the reference window, and the spatial structure-based frame attention weight adjustment of each feature value of the multimodal property fusion characterization feature map of the oily waste is realized to ensure the spatial transformation invariance of the multimodal property fusion characterization feature map of the oily waste under the interaction of feature space, thereby improving the convergence and generalization effect of the decoding regression of the feature set of the multimodal property fusion characterization feature map of the oily waste under the complex spatial structure representation, and improving the accuracy of the control result obtained by inputting it into the crusher speed controller based on the decoder. In this way, the crushing parameters can be adjusted in time according to the specific characteristics of the current material (such as oil content, viscosity and hardness) and the state change of the waste, so as to ensure that the crushing process is more efficient and reliable, avoid the problem of incomplete crushing or excessive crushing, and significantly improve the flexibility and intelligence of the crushing treatment of oily waste.

[0052] In summary, the crushing and processing method of oily waste based on the embodiment of the present application is explained, which uses data analysis and image processing technology based on deep learning to embed and encode the property parameter data such as oil content, viscosity and hardness of the dried oily waste, and extracts multi-scale state features and performs significant joint perception fusion on the detection state image of the dried oily waste, so as to automatically obtain the recommended value of the crushing speed based on the cross-modal fusion characterization features between the embedded coding features of the property parameter data and the multi-scale fusion features of the detection state image. In this way, the crushing parameters can be adjusted in time according to the specific characteristics of the current material and the state changes of the waste, thereby improving the flexibility and intelligence of the crushing and processing of oily waste.

[0053] In other embodiments of the present application, a method for crushing and treating oily waste is also provided, which first conducts a detailed characteristic analysis of the underground oily waste, that is, through a combination of laboratory testing and on-site sampling, obtains key data such as the type, oil content, viscosity, hardness, etc. of the waste. Then, based on these data, in-depth analysis is performed using data analysis software to provide basic data support for subsequent crushing and transportation process design; finally, based on the characteristic analysis, suitable crushing machine models are selected for different types of oily waste, and the optimal crushing method and parameters are determined through experiments. In this way, through continuous optimization and adjustment, rapid and thorough crushing of waste can be achieved to improve processing efficiency.

Claims

1. A method for crushing and treating oily waste, characterized in that: include: Prepare oily waste; dry the oily waste to obtain dried oily waste; transport the dried oily waste to a crushing device for crushing to obtain crushed oily waste; wherein, transport the dried oily waste to a crushing device for crushing to obtain crushed oily waste, including: obtaining property parameter data of the dried oily waste, wherein the property parameter data includes oil content, viscosity and hardness; obtaining a detection state image of the dried oily waste; passing the detection state image of the dried oily waste through an oily waste state feature extractor to obtain a shallow state feature map of the oily waste and a semantic state feature map of the oily waste; The shallow state feature map of the oily waste and the semantic state feature map of the oily waste are subjected to a salient feature joint perception module based on an attention mechanism to obtain a multi-scale joint perception fusion feature map of the oily waste state; the property parameter data of the dried oily waste are embedded and encoded to obtain an oily waste property embedding coding vector; the oily waste property embedding coding vector and the multi-scale joint perception fusion feature map of the oily waste state are subjected to a Transformer-like structure encoder based on cross-modal feature fusion to obtain a multi-modal property fusion representation feature map of the oily waste; based on the multi-modal property fusion representation feature map of the oily waste, a control result is obtained, and the control result is used to represent a recommended value of the crushing rotation speed; The method comprises: performing feature shape reshaping on the shallow state feature map of the oiled waste and the semantic state feature map of the oiled waste to obtain a shallow state feature shape reshaping matrix of the oiled waste and a semantic state feature shape reshaping matrix of the oiled waste; inputting the shallow state feature shape reshaping matrix of the oiled waste and the semantic state feature shape reshaping matrix of the oiled waste into a feature channel-by-channel interactive perception module to obtain a shallow-semantic dependency matrix of the oiled waste and a semantic-shallow dependency matrix of the oiled waste; inputting the shallow-semantic dependency matrix of the oiled waste and the semantic-shallow dependency matrix of the oiled waste into a random dropout module to obtain a pruned shallow-semantic dependency matrix of the oiled waste; and obtaining a plurality of fusion feature maps of the oiled waste state. The invention relates to a method for obtaining a multi-scale joint perception fusion feature map of oiled waste state by multiplying the pruned oiled waste shallow-semantic dependency matrix and the oiled waste semantic state feature shape reshaping matrix to obtain a dependency-optimized oiled waste semantic state feature matrix; the pruned oiled waste semantic-shallow dependency matrix and the oiled waste shallow state feature shape reshaping matrix are matrix-multiplied to obtain a dependency-optimized oiled waste shallow state feature matrix; the dependency-optimized oiled waste shallow state feature matrix and the dependency-optimized oiled waste semantic state feature matrix are feature-reshaped to obtain an optimized oiled waste shallow state feature map and an optimized oiled waste semantic state feature map; the weighted sum of the optimized oiled waste shallow state feature map and the optimized oiled waste semantic state feature map is calculated to obtain the oiled waste state multi-scale joint perception fusion feature map.

2. The method for crushing and treating oily waste according to claim 1, characterized in that: The detection state image of the dried oily waste is passed through an oily waste state feature extractor to obtain a shallow state feature map of the oily waste and a semantic state feature map of the oily waste, including: passing the detection state image of the dried oily waste through an oily waste state feature extractor based on a dilute pyramid network to obtain the shallow state feature map of the oily waste and the semantic state feature map of the oily waste.

3. The method for crushing and treating oily waste according to claim 2, characterized in that: The shallow state feature shape reshaping matrix of the oiled waste and the semantic state feature shape reshaping matrix of the oiled waste are input into the feature channel-by-channel interactive perception module to obtain the shallow-semantic dependency matrix of the oiled waste and the semantic-shallow dependency matrix of the oiled waste, including: calculating the matrix product between the transposed matrix of the shallow state feature shape reshaping matrix of the oiled waste and the shape reshaping matrix of the semantic state feature of the oiled waste to obtain a first correlation feature matrix; calculating the positional division of the scale of the first correlation feature matrix and the first correlation feature matrix to obtain a first correlation feature modulation matrix; calculating the matrix product between the transposed matrix of the semantic state feature shape reshaping matrix of the oiled waste and the shallow state feature shape reshaping matrix of the oiled waste to obtain a second correlation feature matrix; calculating the positional division of the scale of the second correlation feature matrix and the second correlation feature matrix to obtain a second correlation feature modulation matrix; and inputting the first correlation feature modulation matrix and the second correlation feature modulation matrix into the Softmax function respectively to obtain the shallow-semantic dependency matrix of the oiled waste and the semantic-shallow dependency matrix of the oiled waste.

4. The method for crushing and treating oily waste according to claim 3, characterized in that: The weighted sum of the optimized shallow state feature map of oiled waste and the optimized semantic state feature map of oiled waste is calculated to obtain the multi-scale joint perception fusion feature map of the oiled waste state, including: multiplying the optimized shallow state feature map of oiled waste and the optimized semantic state feature map of oiled waste by corresponding weight hyperparameters by position, and then adding the obtained optimized weighted shallow state feature map of oiled waste and the optimized weighted semantic state feature map of oiled waste by position to obtain the multi-scale joint perception fusion feature map of the oiled waste state.

5. The method for crushing and treating oily waste according to claim 4, characterized in that: Embedding the property parameter data of the dried oily waste to obtain the oily waste property embedded coding vector, including: embedding the property parameter data of the dried oily waste using a property parameter embedded coding matrix to obtain the oily waste property embedded coding vector.

6. The method for crushing and treating oily waste according to claim 5, characterized in that: The oil-stained waste property embedding coding vector and the oil-stained waste state multi-scale joint perception fusion feature map are passed through a Transformer-like structure encoder based on cross-modal feature fusion to obtain a multi-modal property fusion representation feature map of oil-stained waste, including: performing layer normalization and convolution coding processing on the multi-scale joint perception fusion feature map of the oil-stained waste state to obtain a multi-scale joint perception space deepening feature map of the oil-stained waste state; copying the multi-scale joint perception space deepening feature map of the oil-stained waste state to obtain a backup multi-scale joint perception space deepening feature map of the oil-stained waste state; and reshaping the backup multi-scale joint perception space deepening feature map of the oil-stained waste state and the multi-scale joint perception space deepening feature map of the oil-stained waste state to obtain a backup multi-scale joint perception space deepening feature matrix of the oil-stained waste state and a multi-scale joint perception space deepening feature matrix of the oil-stained waste state. as the simulated query feature matrix and the simulated key feature matrix; calculate the product of the simulated query feature matrix and the transposed matrix of the simulated key feature matrix to obtain the multi-scale joint perception local autocorrelation library matrix of the oiled waste state; input the multi-scale joint perception local autocorrelation library matrix of the oiled waste state into the attention gating unit based on the GELU function to obtain the multi-scale joint perception gated local autocorrelation library weight matrix of the oiled waste state; use the oiled waste property embedding coding vector as the value feature vector, calculate the product between the multi-scale joint perception gated local autocorrelation library weight matrix of the oiled waste state and the oiled waste property embedding coding vector to obtain the oiled waste property conditional weight vector; use the oiled waste property conditional weight vector as the weight vector, perform channel-by-channel weighting on the multi-scale joint perception fusion feature map of the oiled waste state to obtain the multi-modal property fusion representation feature map of the oiled waste.

7. The method for crushing and treating oily waste according to claim 6, characterized in that: The multi-scale joint perception fusion feature map of the state of oiled waste is subjected to layer normalization and convolution coding processing to obtain a multi-scale joint perception spatial deepening feature map of the state of oiled waste, including: performing layer normalization processing on the multi-scale joint perception fusion feature map of the state of oiled waste to obtain a layer-normalized multi-scale joint perception fusion feature map of the state of oiled waste; performing point convolution coding and spatial convolution coding on the layer-normalized multi-scale joint perception fusion feature map of the state of oiled waste to obtain the multi-scale joint perception spatial deepening feature map of the state of oiled waste.

8. The method for crushing and treating oily waste according to claim 7, characterized in that: Based on the multi-modal property fusion characterization feature map of the oily waste, a control result is obtained, including: passing the multi-modal property fusion characterization feature map of the oily waste through a crusher speed controller based on a decoder to obtain the control result.

Images