Kitchen garbage volume and type estimation method based on picture OCR analysis technology

Abstract

The invention provides a kitchen waste volume and type estimation method based on a picture OCR analysis technology, and relates to the technical field of kitchen waste estimation methods. The kitchen garbage volume and type estimation method based on the picture OCR analysis technology comprises the following steps: S1, identifying a kitchen garbage can and materials in the can by using an on-site kitchen garbage can boundary, specification and color identification model based on deep learning and the OCR technology; s2, identifying material types by using a material type and boundary identification model based on deep learning; and S3, estimating the material volume by adopting a volume estimation prediction model based on a CNN network model. Through the vector classification based on YOLO and the CNN convolutional neural network analysis technology, the receiving and transporting operation condition is monitored in real time, the material quantity and barrel volume recognition rate and accuracy of each receiving and transporting shop point are approved from nothing to existence, the receiving and transporting vehicle resources are effectively optimized, oil consumption is reduced, funds are saved, and the kitchen waste harmless treatment efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of kitchen waste estimation methods, and in particular to a method for estimating the volume and type of kitchen waste based on image OCR analysis technology. Background Art

[0002] In the urban food waste collection and transportation system, the method of estimating the volume and type of food waste refers to the method of quantitatively calculating the volume of food waste in the food waste bin through a series of technical means, and distinguishing and classifying its type. This is crucial for the collection and transportation management of food waste, and helps to accurately grasp the collection and transportation situation and rationally plan resources.

[0003] Although some software in the current collection and transportation industry has the function of collection and transportation weighing, there are obvious shortcomings. In terms of technical functions, it is impossible to automatically take a photo of the current food waste barrel and identify the volume and material type of the food waste in the barrel, making it difficult to ensure the authenticity and comprehensiveness of the collection and transportation data. In terms of operating conditions, the collection and transportation vehicles rely on battery power and 4G network when they are parked and turned off. This dependence limits the flexibility and stability of the operation. For example, when the battery is low or the 4G network signal is poor, it may affect the normal collection and transportation operations and data transmission. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the prior art, the present invention provides a method for estimating the volume and type of kitchen waste based on image OCR analysis technology, which solves the problem of difficulty in ensuring the authenticity and comprehensiveness of collection and transportation data, and the problem that the normal collection and transportation operations and data transmission may be affected when the battery power is low or the 4G network signal is poor.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for estimating the volume and type of kitchen waste based on image OCR analysis technology, specifically comprising the following steps:

[0008] S1. Use the on-site kitchen waste bin boundary, specification, and color recognition model based on deep learning and OCR technology to identify the kitchen waste bin and the materials in the bin;

[0009] S2. Use the material type and boundary recognition model based on deep learning to identify the material type;

[0010] S3. Use a volume estimation prediction model based on a CNN network model to estimate the volume of the material. Preferably, the kitchen bucket material recognition model based on deep learning and OCR includes the following steps:

[0011] S101. Data collection: Collect different types of kitchen bucket pictures, including pictures taken at roughly the same angle and under different lighting conditions. At the same time, provide basic materials for training the OCR model:

[0012] S102, data preprocessing and labeling: preprocessing the images collected in S101, including denoising, graying, and resizing, and then labeling the kitchen waste bucket target in the image to prepare for training the OCR model;

[0013] S103, training the OCR model: using the images preprocessed and annotated in S102 for training, the model can be made to recognize kitchen buckets of different sizes (120L, 240L) and kitchen buckets of different colors (green, red, blue-gray). During the training process, the model performance can be optimized by adjusting the model parameters, increasing the amount of training targets, etc.

[0014] Preferably, the YOLO-based training model includes the following steps:

[0015] S201, for the image processed in step S102, obtain the label boundary value based on the annotation file:

[0016] anchors:

[0017] -[48,75,65,120,86,64]#P3 / 8

[0018] -[88,182,99,133,116,252]#P4 / 16

[0019] -[136,177,186,381,186,218]#P5 / 32

[0020] The threshold segmentation method is used to obtain a binary image and estimate the contour;

[0021] S202, feature extraction, dividing the image into multiple grids, calculating the pigment value of the grids, matching the detection target to obtain the location of the kitchen waste bin, the detection method formula is:

[0022] b x =σ(t x )+c x

[0023] b y =σ(t y )+c y

[0024] b w =p w e tw

[0025] b w =pw e tw

[0026] Pr(object)*IOU(b,object)=σ(t o ).

[0027] (III) Beneficial effects

[0028] The present invention provides a method for estimating the volume and type of kitchen waste based on image OCR analysis technology. It has the following beneficial effects:

[0029] 1. The present invention provides a method for estimating the volume and type of kitchen waste based on image OCR analysis technology. Based on YOLO's vector classification and CNN convolutional neural network analysis technology, it can achieve accurate values ​​of the material capacity of kitchen waste barrels for collection and transportation operations, and can grasp the kitchen waste production of each collection and transportation store. By continuously collecting barrel images with recognition errors and adjusting parameters to train new models, after training every 500 barrel images, the barrel volume recognition rate increased by 40-50% month-on-month and the accuracy rate increased by 40-70%, which plays a key role in optimizing the resources of collection and transportation vehicles, effectively reducing the fuel consumption of operating vehicles, and saving money.

[0030] 2. The present invention provides a method for estimating the volume and type of kitchen waste based on image OCR analysis technology. The material type data of each collection and transportation point is accumulated by using the material type analysis technology based on OCR. By continuously collecting barrel material images with identification type errors, adding and adjusting parameters to train new models, every 500 images participating in the training can increase the material type judgment rate by about 10% month-on-month, providing a strong data basis for the reallocation of collection and transportation vehicles, and significantly improving the efficiency of harmless treatment of kitchen waste returned to the factory. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the application flow of the recognition software of the present invention;

[0032] Figure 2 This is a schematic diagram of the effect before the present invention is applied;

[0033] Figure 3 This is a schematic diagram of the effect after the application of the present invention;

[0034] Figure 4 This is a binary image of the kitchen waste bin position obtained by matching the detection target of the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] like Figure 1 As shown, a method for estimating the volume and type of kitchen waste based on image OCR analysis technology in the present invention specifically includes the following steps:

[0037] S1. Use the on-site kitchen waste bin boundary, specification, and color recognition model based on deep learning and OCR technology to identify the kitchen waste bin and the materials in the bin. The kitchen waste bin material recognition model based on deep learning and OCR includes the following steps:

[0038] S101. Data collection: Collect different types of kitchen bucket pictures, including pictures taken at roughly the same angle and under different lighting conditions. At the same time, provide basic materials for training the OCR model:

[0039] S102, data preprocessing and labeling: preprocessing the images collected in S101, including denoising, graying, and resizing, and then labeling the kitchen waste bucket target in the image to prepare for training the OCR model;

[0040] S103, training the OCR model: using the images preprocessed and annotated in S102 for training, the model can be made to recognize kitchen and dining barrels of different sizes (120L, 240L) and different colors (green, red, blue-gray). During the training process, the model performance can be optimized by adjusting the model parameters, increasing the amount of training targets, etc.

[0041] S2. Use the material type and boundary recognition model based on deep learning to identify the material type;

[0042] S3. Use the volume estimation prediction model based on the CNN network model to estimate the material volume.

[0043] The following steps are involved in training the model based on YOLO:

[0044] S201, for the image processed in step S102, obtain the label boundary value based on the annotation file:

[0045] anchors:

[0046] -[48,75,65,120,86,64]#P3 / 8

[0047] -[88,182,99,133,116,252]#P4 / 16

[0048] -[136,177,186,381,186,218]#P5 / 32

[0049] The threshold segmentation method is used to obtain a binary image and estimate the contour;

[0050] S202, feature extraction, divide the image into multiple grids, calculate the pigment value of the grid, match the detection target to obtain the location of the kitchen waste bin, the detection method formula is as follows: Figure 4 As shown:

[0051] b x =σ(t x )+c x

[0052] b y =σ(t y )+c y

[0053] b w =p w e tw

[0054] b w =p w e tw

[0055] Pr(object)*IOU(b,object)=σ(t o )

[0056] Specifically, in the above specific embodiment, the method identifies the software application process as follows:

[0057] Data collection: Collect different types of kitchen bucket images, covering samples at roughly the same angle and different lighting conditions. At the same time, collect basic materials for training the OCR model. These materials will serve as the basic data for subsequent model training to ensure that the model can adapt to the image characteristics of kitchen buckets in various actual scenarios.

[0058] Training the model

[0059] The collected images were preprocessed by denoising, graying, resizing, and labeling the kitchen waste bucket targets in preparation for training the OCR model. The preprocessed and labeled images were used to train the kitchen waste bucket material recognition model based on deep learning and OCR technology, enabling it to recognize kitchen waste buckets of different specifications (such as 120L, 240L) and colors (green, red, blue-gray). During the training process, the model performance was optimized by adjusting model parameters and increasing the number of training samples.

[0060] The label boundary value of the processed image is obtained based on the annotation file, and the threshold segmentation method is used to obtain the estimated contour of the binary image, and feature extraction is performed. The image is divided into multiple grids to calculate the pigment value of the grid, match the detection target to obtain the location of the kitchen waste bin, and complete the relevant calculations according to the specific detection method formula, thereby constructing a training model based on YOLO and a volume estimation prediction model based on the CNN network model for the identification and estimation of material types and volumes.

[0061] Model used: In the actual food waste barrel collection and transportation scenario, when the barrel rises, a video camera is triggered, and the food waste barrel and the materials in the barrel are identified through the trained model, including the use of material type and boundary recognition models based on deep learning to identify the material type, and the use of a volume estimation prediction model based on a CNN network model to estimate the material volume. The data is then sent to the cloud management background to achieve real-time monitoring and data management of the food waste barrel collection and transportation situation.

[0062] Example 1: The recognition result before the application of this method is as follows: Figure 2 Shown

[0063] Before the application of the food waste bucket volume recognition software, the collection and transportation operation could only obtain the total weight information of the collected materials. The food waste production volume of a single restaurant could not be accurately known, and it was often the case that a vehicle only received half a bucket of materials after traveling a long distance, resulting in low collection and transportation efficiency and unreasonable resource allocation. At the same time, since the type of material in the food waste bucket cannot be identified, it is difficult to take classification management measures for different material situations, which is not conducive to the subsequent optimization of the food waste harmless treatment process.

[0064] Example 2: The recognition result after application of this method is as follows Figure 3 Shown

[0065] After applying the software, the vector classification based on YOLO and the CNN convolutional neural network analysis technology come into play, which can accurately grasp the production volume of kitchen waste at each collection and transportation store, and realize the accurate numerical acquisition of the material capacity of the kitchen waste barrel for collection and transportation operations. After continuously collecting barrel images with recognition errors and adjusting parameters to train new models, after applying the training model for every 500 barrel images, the barrel volume recognition rate increased by 40-50% month-on-month, and the accuracy rate increased by 40-70%. In terms of material type recognition, based on the OCR material type analysis technology, the material type data of each collection and transportation point is accumulated. Every 500 images participating in the training model can increase the material type judgment rate by about 10% month-on-month, providing a reliable data basis for the reallocation of collection and transportation vehicles, greatly improving the efficiency of harmless treatment of kitchen waste returned to the factory, and effectively optimizing the operating efficiency of the entire kitchen waste collection and transportation system.

[0066] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for estimating the volume and type of kitchen waste based on image OCR analysis technology, characterized in that: The specific steps include: S1. Use the on-site kitchen waste bin boundary, specification, and color recognition model based on deep learning and OCR technology to identify the kitchen waste bin and the materials in the bin; S2. Use the material type and boundary recognition model based on deep learning to identify the material type; S3. Use the volume estimation prediction model based on the CNN network model to estimate the material volume.

2. The method for estimating the volume and type of kitchen waste based on image OCR analysis technology according to claim 1, characterized in that: The kitchen bucket material recognition model based on deep learning and OCR includes the following steps: S101. Data collection: Collect different types of kitchen bucket pictures, including pictures taken at roughly the same angle and under different lighting conditions. At the same time, provide basic materials for training the OCR model: S102, data preprocessing and labeling: preprocessing the images collected in S101, including denoising, graying, and resizing, and then labeling the kitchen waste bucket target in the image to prepare for training the OCR model; S103, training the OCR model: using the images preprocessed and annotated in S102 for training, the model can be made to recognize kitchen buckets of different sizes (120L, 240L) and kitchen buckets of different colors (green, red, blue-gray). During the training process, the model performance can be optimized by adjusting the model parameters, increasing the amount of training targets, etc.

3. The method for estimating the volume and type of kitchen waste based on image OCR analysis technology according to claim 1, characterized in that: The following steps are involved in training the model based on YOLO: S201, for the image processed in step S102, obtain the label boundary value based on the annotation file: anchors: -[48,75,65,120,86,64]#P3 / 8 -[88,182,99,133,116,252]#P4 / 16 -[136,177,186,381,186,218]#P5 / 32 The threshold segmentation method is used to obtain a binary image and estimate the contour; S202, feature extraction, dividing the image into multiple grids, calculating the pigment value of the grids, matching the detection target to obtain the location of the kitchen waste bin, the detection method formula is: b x =σ(t x )+c x b y =σ(t y )+c y b w =p w e tw b w =p w e tw Pr(object)*IOU(b,object)=σ(t o )。

Images