Method for analysing waste

Abstract

A method for analysing waste, wherein the method comprises an image creation process (1) during which at least one image of at least one waste bale segment (23) is created, is characterized in that the method comprises a bale instance creation process (2) during which, for each waste bale segment (23) present in the image, a respective bale instance (24) is created.

Description

[0001] Title:

[0002] Method for analysing waste

[0003] Technical Field

[0004] The invention relates to a method for analysing waste according to the preamble of claim 1 . The invention furthermore relates to a method for training machine learning classifiers, an image creation system, a bale instance creation system, a bale instance analysis system, a waste bale analysis system, a computer program and a computer-readable medium.

[0005] Background Art

[0006] Bales are commonly used in the waste industry to transport material and are an essential part of the value chain. However, a guality assessment and thus estimation of market and process potential of individual bales relies heavily on the subjective view of humans so far. In particular, in recycling plants, humans are typically carrying out visual inspections of arriving waste bales and are manually classifying them for future use, for example in terms of materials comprises in the bale and in terms of guality of these materials.

[0007] The industry therefore heavily relies on human-based visual assessment of waste bales, which do not provide guantifiable, documentable, and repeatable information about the bale. Some solutions for automating these tasks, for example using cameras, have been suggested in the past. For example, WO2017 / 207610A1 relates to a method and an apparatus for analysing inhomogeneous bulk cargo. In general, the available solutions do, however, not provide analysis of bales at large varieties, meaning different types of bales (cardboard, plastics, metals, etc.), at a large scale (hundreds or thousands of bales per year and system) and with sufficient preciseness. Furthermore, such solutions are often technically complex and for example require a multitude of sensors.

[0008] Disclosure of the Invention

[0009] It is the problem of the invention to overcome or to at least diminish the above-mentioned disadvantages.

[0010] The problem is solved by a method for analysing waste, wherein the method comprises an image creation process during which at least one image of at least one waste bale segment is created, wherein the method comprises a bale instance creation process during which, for each waste bale segment present in the image, a respective bale instance is created.

[0011] The problem is solved by this method because, by systematically creating images of waste bale segments and subsequently systematically creating bale instances for each waste bale segment, it becomes possible to track and analyse waste of different types that is being fed into a process, such as a recycling process, in a simple, efficient and precise manner. This systematic use of image data and systematic creation of bale instances also makes it possible to rely the waste analysis strongly on visual data, thereby reducing the necessity for complex sensor systems and thus increasing the simplicity of the waste analysis. Furthermore, the simplicity, efficiency and preciseness make it possible to apply the method to a large quantity of waste bales, in particular because the systematic creation of waste bale instances helps the further waste analysis and the tracking of waste bales. In this regard the term “analysing” is to be understood broadly, for example such that it comprises or means determining at least one property of the waste, in particular, one property of a waste bale. The term “instance” shall in particular be understood as relating to a computerised representation of a bale segment or a latent representation of bale segment. The expression “bale segment” itself is to be understood such that each image and / or bale instance does not necessarily have to comprise entire bales; it is rather also acceptable to use images and / or bale instances that comprise only parts or segments of waste bales, for example in cases where waste bales have fallen apart. A bale segment can therefore comprise or be part of one bale. Furthermore, a bale segment can comprise or be one bale. Furthermore, a bale segment can comprise or be one bale and some surroundings, for example some parts of a background of a bale. Furthermore, a bale segment can comprise or be a multitude of bales, possibly with some surroundings. Furthermore, a bale segment can comprise or be composed of parts of a multitude of bales. In the following, a “waste bale” is sometimes also referred to simply as “bale”. In other words, the expressions “waste bale” and “bale” are used as synonyms.

[0012] In a typical embodiment, during the image creation process, the image is created by means of a camera system, wherein the camera system comprises at least one stationary camera and / or at least one movable camera and / or at least one handheld camera and / or at least one vehicle-mounted camera. In a preferred embodiment, the stationary camera is installed at a conveyor belt or in the vicinity of a conveyor belt and overlooking the conveyor belt, at least partly, or at least such that it can create images of objects and in particular bales moving on this conveyor belt. In preferred embodiments, the handheld camera is a camera of a mobile phone or a camera of another handheld or wearable device. In particular embodiments, the movable camera is a drone camera. In some preferred embodiments, the vehicle-mounted camera is a forklift-mounted camera.

[0013] In typical embodiments, the image is created such that a layer structure is visible in the image and / or in that the image comprises multiple side views of one and the same bale. In typical embodiments, the image creation process comprises creating a video stream and extracting single images from this stream. In typical embodiments, the image creation process comprises creating multiple images and / or multiple views of one and the same bale. In this regard, the expression “layer structure" is to be understood such that it refers to a layer structure that looks like a sandwich layer structure, typically comprising a multitude of layers of different types of waste. However, it is not absolutely necessary for the images to comprise a visible layer structure. A method rather also works if images are created such that no layer structures are visible, for example by taking the images from a side of a bale where no layer structure is visible. However, the inventors have found by means of experiments that it can help during a subsequent image analysis process of the bale instances and that it can for example make the waste analysis more precise if a layer structure is visible in the image.

[0014] In typical embodiments, during the bale instance creation process, a neural network architecture identifies each waste bale segment present in the image, wherein the neural network architecture preferably comprises convolutional layers and / or fully connected layers and / or transformer layers, wherein the neural network architecture is preferably used for objected detection. In typical embodiments, this neural network architecture preferably comprises at least one convolutional layer and / or at least one fully connected layer and / or at least one transformer layer.

[0015] In typical embodiments, the bale instance creation process comprises a cropping routine and / or a bale outline detection routine. The inventors have found that such cropping routines or bale outline detection routines help in creating bale instances that can later be analysed in an efficient and a reliable manner. In typical embodiments, the cropping routine comprises a cropping to a bounding box or to bounding boxes and / or a segmenting and / or a removing of a background of an image. In typical embodiments, also the bale outline detection routine comprises such a cropping and / or such a segmenting and / or such a removing of a background. In typical embodiments, the bale outline detection routine comprises detecting an outline of a bale and preferably a drawing of a contour line between waste and background. The cropping routine and / or the bale outline detection routine are typically implemented by means of automated computer algorithms. Instead of using such algorithms for bale section detection, it is also possible that a fixed mask, for example shown in the screen of a camera to a user, is used. A user then typically takes an image using this fixed mask, by placing the fixed mask shown in the screen onto the bale segment of which a bale instance is to be created. Instead of a fixed mask, it is also possible to provide a user-defined box which the user can use to place the bale segment in the correct area of the screen. In typical embodiments, the user draws a bounding box around a desired bale segment immediately after creating the image.

[0016] In preferred embodiments, the method comprises an object tracking process for tracking moving bales, wherein the object tracking process is preferably configured to track bale instances over a sequence of images. If the bales are or the camera is moving, then an optional object tracker tracks the detected bale instances over a sequence of images by assigning a unique (permanent or temporary) tracking ID to each bale (e.g. when filming over a period of time). This is typically done by comparing the bale instances to check for similarity and / or by extracting and extrapolating features using optical flow analysis and / or or by predicting the location of a next detection of a particular bale instance and determining relative motion with a unique ID, typically in order to avoid identifying the same bales twice.

[0017] In typical embodiments, the method comprises a feature vector creation process during which, for each bale instance, a unique feature vector is created, preferably by passing each bale instance through a neural network or a series of neural networks, wherein the neural network or the series of neural networks preferable comprise(s) one or more convolutional layer(s) and / or one or more fully-connected layer(s) and / or one or more transformer layer(s) and / or one or more recurrent layer(s). The inventors have found that such a feature vector creation process can be used to create a unique feature vector for each bale instance, which can be compared to a “fingerprint” of a bale instance. Creating such fingerprints then makes it possible to track and / or trace each bale instance during the subsequent analysing steps of the method in a precise and reliable way. In typical embodiments, multiple neural networks run in parallel during the feature vector creation process. Such fingerprints can for example be used to trace a bale through an entire recycling chain. A feature vector can also be described as tensor.

[0018] In typical embodiments, the method comprises a bale instance analysis process, during which a bale parameter set is determined for each bale instance. Such a bale parameter set has the advantage of grouping parameters for each bale instance in a standardised manner, thereby for example simplifying and standardizing further analysis. One objective of such a waste bale analysis by means of the bale instance analysis process is to determine bale parameters which are essential for the proper handling, evaluation and processing of a waste bale during the recycling process. These generally depend on the target material, but include elements such as material type, composition, quality, color, degree of contamination, and wetness, among others. For example, for plastics, detecting foreign contamination (such as PET bottles in an HDPE bale) is crucial as it can chemically affect the process if not detected and mitigated while for cardboard and paper, color distribution (how white or brown the bale is), foreign objects, as well as moisture are more important, as they have an influence of output product potential and make it possible to estimate the fiber content of a bale.

[0019] In typical embodiments, the bale parameter set comprises a material type parameter and / or a market standard material category parameter and / or a colour composition parameter and / or a homogeneity parameter and / or a composition of material parameter and / or a producer parameter and / or a brand parameter and / or a degree of dirtiness parameter and / or a contaminant parameter and / or an unwanted material parameter and / or a moisture parameter a height parameter, and / or a width parameter, and / or a depth parameter, and / or a volume parameter, and / or a weight parameter and / or a smell parameter and / or a material quality parameter and / or an inventory parameter and / or a subjective quality assessment parameter, wherein the subjective quality assessment parameter preferably comprises a rating between 0 / 10 and 10 / 10, and / or a composition of individual objects parameter and / or a letter-based quality rating, comprising for example the letters “A”, “B” and “C”, wherein “A” indicates highest quality, “B” indicates medium quality and “C” indicates lowest quality. In typical embodiments, the parameter set comprises one or more of at least some of these parameters. In typical embodiments, each of these parameters can also include multiple values and / or multiple objects. In such cases, it is also possible to refer to these parameters as parameters lists. The expression “composition of objects parameter” is to be understood such that it describes a parameter that comprises information about one or more individual objects present in the bale segment, wherein this information is typically fed into the instance analysis process as external data meaning that the individual objects can for example be determined by an object detection method which is not necessarily part of the method according to the invention but which is carried out externally. In typical embodiments, the method according to the invention has an interface for exchanging data with other methods, for example object detection methods for detecting individual objects in waste bales. The expression “interface” (also when mentioned elsewhere in this description) is typically to be understood such that it relates to a virtual interface implemented by means of computer program code. In general, in preferred embodiments, the parameters mentioned above relate to a bale instance and / or or to a waste bale and / or a waste bale segment to which the particular bale instance corresponds.

[0020] In typical embodiments, the bale instance analysis process comprises a classification subprocess during which each bale instance is passed along an analysis pipeline for, step by step, creating the parameter set, wherein the analysis pipeline typically comprises a graph-like structure, wherein the analysis pipeline typically comprises a multitude of classifiers. The expression “pipeline” (also when mentioned elsewhere in this description) is typically to be understood such that it relates to a virtual pipeline implemented by means of computer program code. In typical embodiments, each bale instance is passed along the pipeline, wherein the pipeline preferably comprises an adaptable and / or extendable graph-like structure, such as a decision-tree. At each node of the decision-tree, new attributes correlating to the features of the bale instance are analysed, and preferably more fine-grained categorizations are made at each node. The classifiers comprise of, for example, conditional if-statements and / or machine learning classifiers and / or output analysis from computer vision algorithms. Examples of machine learning classifiers are neural networks.

[0021] In typical embodiments, the bale instance analysis process comprises a feature extraction subprocess during which one or more optical feature(s) is / are extracted from the bale instance, wherein the optical feature(s) comprise(s): one or more colour value(s) and / or one or more contrast value(s) and / or one or more saturation value(s) and / or one or more colour distribution value(s) and / or one or more colour contrast value(s) and / or one or more pixel intensity value(s) and / or one or more pixel saturation value(s) and / or one or more similar optical value(s), wherein the feature extraction subprocess typically supplies the optical feature(s) as input to the classification subprocess, at least partly. In preferred embodiments, the optical features comprise edges and / or edge-orientations and / or keypoints, such as SIFT-features, and / or reflections. In typical embodiments, the parameter set comprises one or more of the above-mentioned optical features. In typical embodiments, the feature extraction subprocess adds one or more or all of the above-mentioned optical features to the parameter set or is at least configured to do so. In typical embodiments, the optical feature^) comprise(s) one or more of the following: an amount and / or a composition of individual objects in the bale instance, such as material distribution including PET, HDPE, foils or similar. Such an amount and / or composition of individual objects is typically determined based on object detection of individual items that are visible on the exterior surface of the bale instance and / or the bale segment.

[0022] In typical embodiments the bale instance analysis process comprises a pre-processing subprocess during which a graphical pre-processing is carried out on the bale instance such that a pre-processed bale instance is created, wherein the preprocessing subprocess preferably comprises an image augmentation and / or an image cropping and / or an image transformation and / or an image rotation and / or an image mirroring and / or a creation of one or more contrast value(s) and / or one or more saturation value(s) and / or a balancing, preferably a white balancing or colour balancing or a combination of both, wherein the pre-processing subprocess typically supplies the pre-processed bale instance as input to the classification subprocess. In typical embodiments, the pre-processing subprocess preferably comprises filtering and / or rotation and / or scaling and / or color transformation and / or cropping. In general, such a pre-processing can have the advantage of making the method more robust to variations from images coming from different sources, such as a phone, static installation, or moving forklift.

[0023] In typical embodiments, the bale instance analysis process comprises an expert knowledge input subprocess, and / or a sensor input subprocess, and / or a user input subprocess. During the expert knowledge input subprocess, a human expert typically has the possibility to feed expert knowledge into the bale instance analysis process which can be subjective expert knowledge or the like. The bale instance analysis process typically comprises a virtual interface for allowing such expert knowledge to be input. During the user input subprocess, a user, which does not necessarily have to be an expert, has the possibility to input information into the bale instance analysis process. During the sensor input subprocess, technical sensors such as optical sensors, moisture sensors, radar sensors, x-ray sensors or other types of sensors typically feed measured values into the bale instance analysis process. In typical embodiments, the bale instance analysis process comprises an interface for accepting information from such sensors and / or a virtual interface for accepting such inputs from users.

[0024] In typical embodiments, the method comprises a reporting process during which the parameter set(s) is / are preferably reported to a customer, wherein the reporting process preferably comprises displaying the parameter set(s) and / or data derived from the parameter set(s) on a display, and / or wherein the reporting process preferably comprises sending the parameter set(s) and / or data derived from the parameter set(s) in an email and / or wherein the reporting process preferably comprises printing the parameter set(s) and / or data derived from the parameter set(s) to an electronic file, in particular a pdf-file or a printable file. In typical embodiments, a sending of the parameter set to a database and optionally a visualisation of data in that database is initiated before the reporting process is started or is running in parallel with the reporting process. In typical embodiments, the reporting process takes data from the database as input.

[0025] In typical embodiments, the method comprises a storage recommendation process, during which, preferably based on the bale parameter set, a recommendation for a storage position inside a stock of waste bales for the bale to which the bale instance corresponds is made, wherein the recommendation process preferably comprises a comparison between the bale parameter set and bale details, in particular bale parameter sets, of a multitude of waste bales in the stock of waste bales. Such a recommendation process has the following advantage: After the bale instance analysis process has been carried out, typically at an entry level of a recycling plant, the information gathered during the recycling process, for example the parameter sets created for the different bale instances or any other data present in the database, can be used to give recommendations of where to place a particular bale in the stock of waste of the recycling plant. For example, after the analysis of a bale has been performed and a clear understanding of the type of this bale has thus been gained, it becomes possible to make a recommendation for a precise storage location for this particular bale inside the stock, typically a storage location where bales of similar types are already stocked. Like this, later on during a recycling process, when bales of a particular type (that is: with a particular content) are needed during the recycling process, it is sufficient to simply take a bale from a particular storage location where bales of this particular type are being stocked. Like this, a recipe of the recycling process can be maintained in a very efficient manner, the recipe can be followed more precisely and the quality of the material leaving the recycling process can be improved.

[0026] All above-mentioned processes, subprocesses, routines and the like can in principle also work in a standalone manner, meaning that in certain embodiments of the invention, they can exist alone and separately from the overall method for analysing waste. Such standalone embodiments are considered to be comprised in the present description and form part of the invention.

[0027] The problem is furthermore solved by a method for training machine learning classifiers for use in a method for analysing waste according to any of the embodiments described above. In such a method for training machine learning classifiers, the machine learning classifiers are trained by one or more of the following:

[0028] - training, preferably in a supervised manner, using labelled data, wherein the labelled data typically comprises bale instances labelled by a human expert, and / or

[0029] - training, preferably in a supervised manner, using synthetically generated data, wherein the synthetically generated data typically comprises synthetically generated bale instances of waste bales or of waste, and / or

[0030] - training in an unsupervised manner, in particular training on finding similarities in different images of one and the same bale and / or in different images of multiple bales, and / or

[0031] - training, preferably in a supervised manner, using labelled data, wherein the labelled data typically comprises bale instances combined with individually detected waste objects used to create, during a bale creation process, the bales to which the bale instances correspond, and / or

[0032] - training, preferably in a supervised manner, using labelled data, wherein the labelled data typically comprises bale instances combined with information gathered during any of the processes and / or subprocesses and / or routines and / or steps of the method for analysing waste, in particular information gathered during the bale instance analysis process, or during a subsequent processing of a bale.

[0033] In this regard, the expression “information gathered during any of the processes and / or subprocesses and / or routines and / or steps of the method for analysing waste, in particular information gathered during the bale instance analysis process, or during a subsequent processing of a bale” is to be understood such that it relates to any information on a bale that has been processed by the method for analysing waste according to the invention by means of analysing one or more instances of this bale and / or on any information gathered during a subsequent processing of such a bale, for example during a recycling process in a recycling plant. Such information can be referred to as “bale information”. Such bale information can, for example, comprise:

[0034] - information gathered during an analysis of flakes after the bale has been shredded or foreign material has been filtered out, and / or

[0035] - information on individual items that have been baled in a bale, for example information on material type and / or brands and / or material composition, and / or

[0036] - type and / or amount of additives that are consumed during subsequent processing of the bale, for example during the recycling of the waste comprised in the waste bale that has been analysed by means of the method for analysing waste, and / or

[0037] - composition of material obtained after the bale has been opened, and / or

[0038] - information of unusable material (also referred to as reject) created after the bale has been opened, and / or

[0039] - information on clogging levels of filters that are used during the method for analysing waste, and / or

[0040] - information on process anomalies encountered during the processing of a bale by means of the method for analysing waste according to the invention or during subsequent processing of such a bale, for example paper tears in case of cardboard, and / or

[0041] - information on the recyclate material properties and quality. In typical embodiments of the method for training machine learning classifiers, any of such bale information (in particular bale information of a multitude of bales) is used to create labelled data, in particular used to label a dataset of bale images. In general, the labelled data comprises, at least in certain embodiments, a dataset of bale images that can be used for training classifiers for use in a method for analysing waste according to the invention.

[0042] In certain embodiments of the invention, the method for training machine learning classifiers can also be part of the method for analysing a waste bale according to the invention. In other words, the method for training machine learning classifiers can be integrated within the method for analysing waste according to the invention, for example as a subprocess or a subroutine.

[0043] In general, the above-mentioned methods are typically computer-implemented methods, at least partly.

[0044] In typical embodiments, the method continuously saves created bale instances as an annotated dataset and continuously uses this dataset to further refine itself, namely the method.

[0045] In typical embodiments, a standardized quality metric and / or a standardized value metric is determined for each waste bale handled by the method, in particular based on the parameter sets.

[0046] The problem is furthermore solved by an image creation system, configured to create at least one image of at least one waste bale segment, wherein the system comprises a camera system, wherein the camera system comprises at least one stationary camera and / or at least one moveable camera and / or at least one handheld camera and / or at least one vehicle-mounted camera. The image creation system is typically configured to participate in a method according to the invention by carrying out the image creation process outlined above, at least partly. For further explanations concerning the camera system of such an image creation system and other technical details of the image creation system, reference is made to the above technical explanations regarding the method for analysing waste, and in particular to the explanations regarding the image creation process.

[0047] A bale instance creation system according to the invention is configured to receive as input an image, preferably an image created by the image creation system previously described, and to create a respective bale instance for each waste bale segment present in the image. The bale instance creation system is typically configured to participate in a method according to any of the embodiments described above by carrying out the bale instance creation process outlined above, at least partly. For that purpose, the bale instance creation system typically comprises adapted modules, typically implemented by software code, such as a cropping module and / or a bale outline detection module and / or a background removal module and / or a neural network module, typically configured for object detection in digital images, preferably based on the use of convolutional layers and / or fully connected layers and / or transformer layers.

[0048] The problem is furthermore solved by bale instance analysis system configured to receive as input a bale instance, preferably a bale instance created by the bale instance creation system as previously described, and to determine a bale parameter set for the bale instance. The bale instance analysis system is typically configured to participate in the method according to the invention by carrying out the bale instance analysis process, at least partly. In typical embodiments, the bale instance analysis system is configured to receive inputs from the expert knowledge input subprocess and / or the sensor input subprocess and / or the user input subprocess. The bale instance analysis system typically comprises a multitude of modules, typically implemented by means of computer program code, which enable the bale instance analysis system to carry out the different steps, subprocesses and the like of the bale instance analysis process. In typical embodiments, the bale instance analysis system comprises for example a classification module and / or feature extraction module and / or a pre-processing module and / or an expert knowledge input module and / or a sensor input module and / or a user input module.

[0049] The problem is furthermore solved by a waste bale analysis system, comprising an image creation system as described above and a bale instance creation system as described above and a bale instance analysis system as described above and preferably a database configured to hold a multitude of bale parameter sets for a multitude of bale instances and preferably a visualization system for visualizing the database, at least partly. The systems are each preferably implemented by means of computer program code, at least partly. In typical embodiments, the waste bale analysis system comprises an object tracking module configured to carry out the above- mentioned object tracking process and / or a feature vector creation module configured to carry out the above-mentioned feature vector creation process. In particular embodiments, the waste bale analysis system comprises a multitude of image creation systems as described above and / or a multitude of bale instance creation systems as described above and / or a multitude of a bale instance analysis systems as described above and / or a multitude of databases configured to hold a multitude of bale parameter sets for a multitude of bale instances and / or a multitude of visualization systems for visualizing the database, at least partly. In particular embodiments, the waste bale analysis system comprises a multitude of scanning systems for tracing waste bales through a recycling process that employs a method according to the invention. In preferred embodiments, the waste bale analysis system comprises a recommendation system comprising means configured to carry out a recommendation process during which, preferably based on the bale parameter set, a recommendation for a storage position inside a stock of waste bales for the bale to which the bale instance corresponds is made, wherein the recommendation process preferably comprises a comparison between the bale parameter set and bale details, in particular bale parameter sets, of a multitude of waste bales in the stock of waste bales.

[0050] In particular embodiments, the method and in particular the bale instance creation process is configured to be interfaced with stock management in order to automatically check-in or check-out bales from a stock inventory or give recommendations for placing in the stock, for example recommendations on where to stock a particular bale in a group of already stocked bales. This can for example be done by positioning one of the above-mentioned systems at a receiving area for bale deliveries in a recycling plant to automatically add bales to the inventory or on the conveyor belt leading to the material processing to remove it from the stock.

[0051] A computer program comprises, in a typical embodiment of the invention, instructions which, when the program is executed by a computer, cause the computer to carry out a method according to any of the above-mentioned embodiments. The expression “computer” is to be understood as referring to any device or structure that is able to execute the instructions. The computer program can also be referred to as computer program product.

[0052] A computer-readable medium comprises, in an embodiment of the invention, computer program code for carrying out a method according to any of the above-mentioned embodiments and / or comprises a computer program according to the above- mentioned embodiment. The expression “computer-readable medium” can be understood as referring in particular but not exclusively to hard disks and / or servers and / or memory sticks and / or flash drives and / or DVDs and / or Blu-ray disks and / or CDs. Furthermore, the expression “computer-readable medium” can also refer to a data stream which is for example established when a computer program and / or a computer program product is downloaded from the internet. Such a computer-readable medium typically has stored thereon a computer program as described above.

[0053] Brief Description of Figures

[0054] In the following, the invention is explained by means of Figures, wherein show:

[0055] Figure 1 : a schematic view of a method for analysing waste according to a first embodiment of the invention, as block diagram,

[0056] Figure 2: a schematic view of a waste bale analysis system according to one embodiment of the invention, as block diagram, and

[0057] Figure 3: a schematic view of a method for analysing waste according to a second embodiment of the invention, as block diagram. Description of Preferred Embodiments

[0058] Figure 1 shows a schematic view of a method for analysing waste according to one embodiment of the invention, as block diagram. In particular, Figure 1 shows a multitude of processes, subprocesses, routines, methods and the like, which are represented as blocks. Arrows are used to indicate that data is being transferred between the different blocks. The arrows indicate that data transfers between the different blocks are in principle possible and do not necessarily mean that data always has to be transmitted between the different blocks following the different arrows. It is furthermore also possible, that data is transferred between different blocks where, mainly for reasons of clarity, no separate arrow has been indicated. The method for analysing waste shown in Figure 1 comprises an image creation process 1 . During the image creation process 1 , an image of a waste bale segment is created. The creation of the image is typically done using a camera system, for example a stationary camera system installed in the vicinity of a conveyor belt on which waste bales are fed into a recycling process or a moveable camera system, such as a smartphone or tablet with a camera or a camera system installed on a forklift vehicle. Other examples for suitable camera systems are:

[0059] - a dedicated mobile device, possibly with custom hardware and optics,

[0060] - a stationary system mounted on a waste baling or waste debaling system,

[0061] - a stationary system or gate at or prior to the unloading or loading point of a truck,

[0062] - a dedicated “scanning station”, in which a bale is placed and automatically photographed by one or multiple cameras,

[0063] - a camera configured to move around the bale, or the bale could rotate on a turntable or similar in order to capture multiple angles.

[0064] Instead of a single image, a sequence of images can be created in certain embodiments during the image creation process 1. Like this, either a stream of single images or a stream of image sequences are typically created during the image creation process, for further processing by the method and in particular for processing through an analysis pipeline. In typical embodiments, an image of a bale is constructed by a sequence of moving pixels (line scan over time) instead of or in addition of a standard image capture. The image created during the image creation process 1 is then transferred to the bale instance creation process 2 during which a bale instance is created for this waste bale segment. The waste bale segment itself and the waste bale instance itself are not shown in Figure 1 . The waste bale instance can be imagined as computerised representation of the bale segment. The bale instance creation process 2 comprises a cropping routine 3a and a bale outline detection routine 3b. Both of these routines can be used during the creation of the bale instance from the bale segment. For example, the cropping routine 3a can crop a bounding box to the image of the waste bale segment and thereby create the bale instance. The bale outline detection routine 3b can for example automatically draw an outline of a bale in the image comprising the bale segment and thereby create the bale instance based on the image of the bale segment. As mentioned before, a bale segment can be a part of a waste bale or can also be an entire waste bale, possibly with surrounding background. The bale instance creation process 2 can pass the bale instance on to different processes, for example to an object tracking process 4 and / or to feature vector creation process 5 and / or to a bale instance analysis process 6. During the object tracking process 4, moving waste bales, for example on a conveyer belt or inside a factory, can be tracked, if needed also over a sequence of images. During the feature vector creation process 5 a unique feature vector can be created for each bale instance, typically by passing each bale instance through a neural network or a series of neural networks. Like that, a vector, typically comprising optical features of each bale instance, is created, which can be used as “fingerprint” to identify each bale instance.

[0065] During the bale instance analysis process 6, a bale parameter set is determined for each bale instance. The bale parameter set itself is not shown in Figure 1 , but can be imagined as a set of parameters or parameter lists comprising for example a material type parameter and / or a colour composition parameter and / or a homogeneity parameter and / or a moisture parameter and / or a volume parameter and / or a weight parameter and / or a smell parameter and / or a contaminant parameter and / or any of the possible parameters outlined before. In the embodiment of the invention shown in Figure 1 , the bale instance analysis process 6 comprises a classification subprocess 7. During this classification subprocess 7, each bale instance is passed along an analysis pipeline for, step by step, creating the parameter set, wherein the analysis pipeline typically comprises a graph-like structure, wherein the analysis pipeline typically comprises a multitude of classifiers. During the classification subprocess 7, each bale instance is typically passed through the graph-like structure, wherein at each level of the graph-like structure, one or more classifiers are used to classify the bale instance according to different criteria, such that, step by step, parameters for the parameter set are determined. The bale instance analysis process furthermore comprises a feature extraction subprocess 8 and a pre-processing subprocess 9. During the feature extraction subprocess 8, one or more optical features are extracted from the bale instance, for example a colour value and / or a contrast value and / or a saturation value. The feature extraction during feature extraction subprocess 8 the can in certain embodiments be done by means of one or more neural networks. These values can also be fed into the classification subprocess 7. The pre-processing subprocess 9 makes it possible to carry out a graphical pre-processing on the bale instance such that a pre-processed bale instance is created. The graphical pre-processing of the pre-processing subprocess 9 can for example comprise an image augmentation and / or an image transformation and / or an image rotation and / or an image mirroring or multitudes of such operations. As can be seen in Figure 1 and as indicated by the arrows, also the pre-processed bale instance created during the pre-processing subprocess 9 can be fed into the classification sub-process 7. The bale instance analysis process 6 furthermore comprises an expert knowledge input subprocess 10, a sensor input subprocess 11 and a user input subprocess 12. These subprocesses 10, 11 , 12 make it possible to feed expert knowledge, sensor inputs from physical sensors and user input into the classification subprocess 7. The parameter set created by the classification subprocess 7 is then sent to the reporting process 13, during which the parameter set for each bale instance is integrated into a database (not explicitly shown in Figure 1). During the reporting process 13, some or all content of the database can also be made available to a user of the method for analysing waste, for example in the form of graphical displays, in the form of emails or in the form of printable files.

[0066] Figure 1 also shows a method for training machine learning classifiers 14. This method is essentially a method using artificial intelligence mechanisms for drawing conclusions from the bale instance analysis process 6 and / or the reporting process 13 and / or the database and / or any other process of the method for waste analysis. Based on this data, the method for training machine learning classifiers can carry out training operations on the bale instance analysis process 6 and in particular on the classification subprocess 7. Such training can comprise a training using labelled data, wherein the labelled data typically comprises bale instances labelled by human experts and / or training using synthetically generated data and / or training on finding similarities in different images, for example different images of one and the same bale or different images of multiple bales, and / or training in an unsupervised manner and / or training using labelled data, typically labelled through sensor data providing a ground truth, for example sensor data from a humidity sensor.

[0067] It shall be understood that the different processes, routines, methods and the like shown in Figure 1 do not necessarily have to be carried out one after the other but typically are carried out in parallel, at least in some preferred embodiments, especially because the method shown in Figure 1 is typically not dealing with only one image and / or one bale segment and / or one bale instance at a time, but is processing in typical embodiments a multitude of images and / or bale segments and / or bale instances in parallel. Furthermore, the method can also be used to post-process data, for example historic image data and / or historic bale instances. For example, it is possible to look back at several months of historic image data and / or historic bale instances by means of the method or any processes, subprocesses, routines or steps of the method.

[0068] In particular embodiments, features extracted from the different bale instances during the method, for example during the feature extraction subprocess 8, such as color density, saturation, contrast, individual objects particularly etc., can be fed into the classification subprocess and / or can be added to the database, in particular to respective parameter sets, and / or can be used directly for visualization and reporting. For this purpose, one or more of the following algorithms can be carried out during the method, for example during the feature extraction subprocess 8:

[0069] - The color composition of a bale is determined using computer vision algorithms. For example, thresholds are defined in the HSV color space for different colouration properties such as percentages of brown, white and coloured cardboard, or degree of colouration and dirtiness of plastic bales; An object detection algorithm is used to detect outliers and potentially harmful or disrupting objects on the surface of a bale instance or a bale segment or an image of a bale.

[0070] In particular embodiments, the expert knowledge comprises one or more of the following: material type, material quality, location of baling, location of debaling, location of transfer, date and / or time, weather data, batch number, license plate of transporting vehicle, delivery weight, brand information and / or producer information and / or inventory information. In typical embodiments, the expert knowledge comprises customization information which is used during bale instance analysis process 6, preferably during the classification subprocess, to customize a sequence of classifiers used during the classification subprocess, in particular to fit a specific deployment setting of a deployment location of which the expert supplying the expert knowledge has knowledge.

[0071] In typical embodiments, during the bale instance analysis process 6, for example the subprocesses 10, 11 , 12, process relevant information from before baling or after debaling can be fed into the bale instance analysis process 6, for example:

[0072] - information on individual items that have been baled (material type, brands, material composition), and / or

[0073] - type and amount of additives that are consumed in further processing, for example during the recycling of the waste comprised in the waste bales(s) handled by means of the method, and / or

[0074] - composition of the material after a bale has been opened, and / or

[0075] - information of unusable material (reject) created after the bale has been opened and processed, and / or

[0076] - information on clogging levels of filters that are used during the method, and / or

[0077] - information on process anomalies (for example paper tears in case of cardboard), and / or

[0078] - information on the recyclate material properties and quality

[0079] In typical embodiments, weight information for each bale instance and / or each bale can be added to the classifiers used during the classification subprocess 7. Such weight information is typically determined either directly such as by individual weighting of bales on dedicated scales or scales mounted on a conveyor belt, or indirectly, such as by estimating the bale volume and density to calculate weight, or measuring electrical power draw on the conveyor belt, or weighting a complete waste delivery of a truck or an entire loaded truck and obtaining an average value based on the total quantity of bales.

[0080] In particular embodiments, during the method, for example during the reporting process 13, the following actions are carried out: for each bale instance, the generated parameter set is sent to a database with a unique ID and possibly including further metadata, such as data acquired or generated during the different subprocesses of the bale instance analysis method 6 and / or other process, subprocesses and routines of a method for analysing waste according to the invention. From this database, the data comprised in the database is used for visualization (for example in applications such as an online dashboard, mobile app or interfaced with existing ERP / CRM software systems) or aggregated for analysis and further refining of the method, for example for training by means of the method for training machine learning classifiers 14. In typical embodiments, the data in the database can be used (as further steps of the method) for one or more of the following:

[0081] - a comprehensive digitization and tracing of the waste bale,

[0082] - a determination of waste bale material type and / or waste bale quality,

[0083] - a determination of other relevant properties of waste bales,

[0084] - a comparison of quality and cost-efficiency for different suppliers of waste bales, in particular for supplier assessment (also referred to as “supplier vetting”), and / or

[0085] - a comparison of the cost-efficiency for different shifts run in a recycling plant where the method is being carried out, for example by means of calculating running cost of goods for a given production recipe.

[0086] In particular embodiments, the method, for example the reporting process 13, comprises one or more of the following:

[0087] - providing content of the database and / or the parameter set(s) on a dashboard and / or in a digital inventorying system and / or in a digital or physical report; and / or - a feedback step, during which feedback on the data in the database is collected from plant workers in a recycling plant where the method is being carried out, such as approval, rejection, or refining of analysed features, wherein such feedback is then subsequently used to further enhance the data in the database, and / or for amending the classifiers used during the classification subprocess 7, and / or during the method for training machine learning classifiers 14; and / or

[0088] - a monitoring subprocess, comprising continuously determining, based on the data comprised in the database, the quality and value of bales being fed into the recycling process on which the method according to the invention is applied, thereby preferably determining the running cost of goods in the recycling process, which typically allows for optimization of the recycling process and comparison of shifts; and / or

[0089] - a supplier information addition step, preferably comprising adding information about the supplier of a particular waste bale and possibly also purchasing price for each bale, thereby enabling an assessment of how cost-effective, qualitative and reliable a supplier is; and / or

[0090] - a recipe determination step, during which information of the type of waste that was fed into the method is used to select the next bale to follow a recycling process recipe.

[0091] In general, by continuously running waste bale detection and waste analysis during a method according to the invention, and in particular material and quality classification in the beginning of material processing of a recycling plant, the ‘recipe’ for the process can be tracked. This allows to check if the right ‘ingredients’ are fed to the process and to estimate the resulting output material.

[0092] In general, a processing of the methods, processes, subprocesses, routines, steps or the like described in this specification and all associated data processing can be executed on a local edge device, such as a smartphone or computer. Alternatively, the processing of the methods, processes, subprocesses, routines, steps or the like described in this specification and all associated data processing can be processed on an external computer, for example using a cloud provider or a server. Figure 2 shows a schematic view of a waste bale analysis system 18 according to one embodiment of the invention, as block diagram. The waste bale analysis system 18 comprises an image creation system 15, a bale instance creation system 16, a bale instance analysis system 17, a database 19, a visualization system 20, a machine learning training system 21 and a recommendation system 22. The image creation system 15 is configured to carry out an image creation process as previously described. The bale instance creation system 16 is configured to carry out a bale instance creation process as previously described. The bale instance analysis system 17 is configured to carry out a bale instance analysis process 6 as previously described. The database 19 comprises a multitude of parameter sets, for all bale instances that have so far been dealt with by the waste bale analysis system 18. The visualization system 20 is configured to display a desired content of the database 19 and / or to create reports in the form of emails, printable files or the like based on the content of the database 19. The machine learning training system 21 is configured to carry out the method for training machine learning classifiers according to the invention. The recommendation system 22 is configured to carry out a recommendation process during which, preferably based on the bale parameter set, a recommendation for a storage position inside a stock of waste bales for the bale to which the bale instance corresponds is made, wherein the recommendation process preferably comprises a comparison between the bale parameter set and bale details, in particular bale parameter sets, of a multitude of waste bales in the stock of waste bales.

[0093] In typical embodiments, the waste bale analysis system 18 and in particular the image creation system 15 comprises at least one light source for constant illumination. In typical embodiments, the waste bale analysis system 18 and in particular the image creation system 15 comprises 6 a near infrared (NIR) sensor, which is typically configured to operate at wavelengths from 780 nm to 2500 nm. Such an NIR sensor can be used during the method according to the invention and in particular during the image creation process for the following purposes:

[0094] - distinction and / or proportion detection of materials in a bale, such as different types of plastics, cardboard and metals, and / or

[0095] - detection of level of moisture of a cardboard bale, and / or

[0096] - detection of an amount of ash present in a cardboard bale, and / or - detection of an amount of lignin present in a bale comprising cardboard, and / or

[0097] - detection of an amount of organic contamination in a bale.

[0098] During a method according to the invention, such an NIR sensor can in particular be employed to look at the reflection of light of different wavelengths in the NIR spectrum. Then, for example by classical clustering and threshold analysis, or by using machine learning algorithms, materials and material properties in waste bales can be classified. In general, instead of using the NIR sensor in the methods or systems of the invention themselves, it can also be used as a development tool (which is then a tool apart, that can, however, still be regarded as being part of the waste bale analysis system according to the invention) to establish high quality ground truth. Data acquired in such a way can then be used to train neural networks in any of the systems according to the invention in a supervised fashion on regular RGB or RGBD imagery.

[0099] In typical embodiments, the waste bale analysis system 18 comprises a depth sensor. Such a depth sensor can be part of the camera system and can for example be or comprise a 3D-camera a lidar sensor, a single point depth sensor, an ultrasonic sensor, an infrared sensor, a radar sensor or any other depth sensor. In a method according to the invention, such a depth sensor can be used to provide additional information, in particular as inputs to the bale instance analysis process. This can: Together with a segmentation provided by the image creation process and / or the bale instance creation process, a volume of a bale can be determined. By further estimating the density of the bale, based on material classifications, the mass of the bale is determined.

[0100] In typical embodiments, the waste bale analysis system 18 comprises a ground penetrating radar configured to provide information on waste bale interiors to a method according to the invention, in particular to the bale instance analysis process. The ground penetrating radar can for example be configured and can be used for the detection of metallic objects in waste bales. In typical embodiments, the waste bale analysis system 18 comprises a moisture sensor configured to provide information on waste bale moisture content to a method according to the invention, in particular to the bale instance analysis process.

[0101] Figure 3 shows a schematic view of a method for analysing waste according to a second embodiment of the invention, as block diagram. In particular, Figure 3 shows a method for analysing waste according to the invention, wherein the method comprises an image creation process 1 and a bale instance creation process 2. The method consistently receives a stream of waste bales 25 as input. The stream of waste bales 25 is shown by a symbolic arrow in Figure 3. During the image creation process 1 of the method, a camera system 26 is used to take pictures (or record for example a video stream) of the incoming stream of waste bales 25. Based on these pictures or video stream, images of waste bale segments 23 are created. In Figure 3, an image with one waste bale segment 23 is shown, which comprises one waste bale and some background. The image with this waste bale segment 23 is then passed on to the bale instance creation process 2. During the bale instance creation process 2, a bales instance 24 is created for the waste bale segment 23. In cases where images comprise a multitude of waste bale segments (such cases are not shown in Figure 3), one respective bale instance is created for each waste bale segment present in the image during the bale instance creation process 2. The bale instances 24 created during the bale instance creation process 2 can then be analysed in more detail by further optional processes of the method for analysing waste according to the invention, for example by means of any of the processes, subprocesses, routines and / or steps outlined above.

[0102] Furthermore, the following claims are hereby incorporated into the Description of Preferred Embodiments, where each claim may stand on its own as a separate embodiment. While each claim may stand on its own as a separate embodiment, it is to be noted that - although a dependent claim may refer in the claims to a specific combination with one or more other claims - other embodiments may also include a combination of the dependent claim with the subject matter of each other dependent or independent claim. Such combinations are proposed herein unless it is stated that a specific combination is not intended. Furthermore, it is intended to include also features of a claim to any other independent claim even if this claim is not directly made dependent to the independent claim.

[0103] It is further to be noted that methods disclosed in the specification or in the claims may be implemented by a device having means for performing each of the respective acts of these methods.

[0104] List of Reference Signs image creation process bale instance creation process a cropping routine b bale outline detection routine object tracking process feature vector creation process bale instance analysis process classification subprocess feature extraction subprocess pre-processing subprocess 0 expert knowledge input subprocess 1 sensor input subprocess 2 user input subprocess 3 reporting process 4 method for training machine learning classifiers5 image creation system 6 bale instance creation system 7 bale instance analysis system 8 waste bale analysis system 9 database 0 visualization system 1 machine learning training system 2 recommendation system 3 waste bale segment 4 bale instance 5 stream of waste bales 6 camera system

Claims

Patent Claims1 . Method for analysing waste, wherein the method comprises an image creation process (1) during which at least one image of at least one waste bale segment (23) is created, characterized in that the method comprises a bale instance creation process (2) during which, for each waste bale segment (23) present in the image, a respective bale instance (24) is created.

2. Method according to claim 1 , characterized in that during the image creation process (1), the image is created by means of a camera system (26), wherein the camera system (26) comprises at least one stationary camera and / or at least one moveable camera and / or at least one handheld camera and / or at least one vehiclemounted camera.

3. Method according to any of the previous claims, characterized in that the image is created such that a layer structure is visible in the image and / or in that the image comprises multiple side views of one and the same bale.

4. Method according to any of the previous claims, characterized in that, during the bale instance creation process (2), a neural network architecture identifies each waste bale segment (23) present in the image, wherein the neural network architecture preferably comprises convolutional layers and / or fully-connected layers and / or transformer layers, wherein the neural network architecture is preferably used for object detection.

5. Method according to any of the previous claims, characterized in that the bale instance creation process (2) comprises a cropping routine (3a) and / or a bale outline detection routine (3b).

6. Method according to any of the previous claims, characterized in that the method comprises an object tracking process (4) for tracking moving bales, wherein the object tracking process (4) is configured to track bale instances (24) over a sequence of images.

7. Method according to any of the previous claims, characterized in that the method comprises a feature vector creation process (5) during which, for each bale instance (24), a unique feature vector is created, preferably by passing each bale instance (24) through a neural network or a series of neural networks, wherein the neural network or the series of neural networks preferably comprise(s) one or more convolutional layer(s) and / or one or more fully-connected layer(s) and / or one or more transformer layer(s) and / or one or more recurrent layer(s).

8. Method according to any of the previous claims, characterized in that the method comprises a bale instance analysis process (6), during which a bale parameter set is determined for each bale instance (24).

9. Method according to claim 8, characterized in that the bale parameter set comprises:- a material type parameter, and / or- a market standard material category parameter, and / or- a colour composition parameter, and / or- a homogeneity parameter, and / or- a composition of material parameter, and / or- a producer parameter, and / or- a brand parameter, and / or- a degree of dirtiness parameter, and / or- a contaminant parameter, and / or- an unwanted material parameter, and / or- a moisture parameter, and / or- a height parameter, and / or- a width parameter, and / or- a depth parameter, and / or- a volume parameter, and / or- a weight parameter, and / or- a smell parameter, and / or- a material quality parameter, and / or- an inventory parameter, and / or- a subjective quality assessment parameter, wherein the subjective quality assessment parameter preferably comprises a rating between 0 / 10 and 10 / 10, and / or- a composition of individual objects parameter.

10. Method according to any of the claims 8 to 9, characterized in that the bale instance analysis process (6) comprises a classification subprocess (7) during which each bale instance (24) is passed along an analysis pipeline for, step by step, creating the parameter set, wherein the analysis pipeline typically comprises a graph-like structure, wherein the analysis pipeline typically comprises a multitude of classifiers.

11. Method according to any of the claims 8 to 10, characterized in that the bale instance analysis process (6) comprises a feature extraction subprocess (8) during which one or more optical feature(s) is / are extracted from the bale instance (24), wherein the optical feature(s) comprise(s): one or more colour value(s) and / or one or more contrast value(s) and / or one or more saturation value(s) and / or one or more colour distribution value(s) and / or one or more colour contrast value(s) and / or one or more pixel intensity value(s) and / or one or more pixel saturation value(s) and / or one or more similar optical value(s), wherein the feature extraction subprocess (8)typically supplies the optical feature(s) as input to the classification subprocess (7), at least partly.

12. Method according to any of the claims 8 to 11 , characterized in that the bale instance analysis process (6) comprises a pre-processing subprocess (9) during which a graphical pre-processing is carried out on the bale instance (24) such that a pre-processed bale instance is created, wherein the pre-processing subprocess (9) preferably comprises an image augmentation and / or an image cropping and / or an image transformation and / or an image rotation and / or an image mirroring and / or a creation of one or more contrast value(s) and / or one or more saturation value(s) and / or a balancing, preferably a white balancing or colour balancing, wherein the pre-processing subprocess (9) typically supplies the pre-processed bale instance as input to the classification subprocess (7).

13. Method according to any of the claims 8 to 12, characterized in that the bale instance analysis process (6) comprises:- an expert knowledge input subprocess (10), and / or- a sensor input subprocess (11), and / or- a user input subprocess (12).

14. Method according to any of the claims 8 to 13, characterized in that the method comprises a reporting process (13) during which the parameter set(s) is / are preferably reported to a customer, wherein the reporting process (13) preferably comprises displaying the parameter set(s) and / or data derived from the parameter set(s) on a display, and / or wherein the reporting process (13) preferably comprises sending the parameter set(s) and / or data derived from the parameter set(s) in an email and / or wherein the reporting process (13) preferably comprises printing the parameter set(s) and / or data derived from the parameter set(s) to an electronic file, in particular a pdf- file or a printable file.

15. Method according to any of the previous claims, characterized in that the method comprises a storage recommendation process, during which, preferably based on the bale parameter set, a recommendation for a storage position inside a stock of waste bales for the bale to which the bale instance (24) corresponds is made, wherein the recommendation process preferably comprises a comparison between the bale parameter set and bale details, in particular bale parameter sets, of a multitude of waste bales in the stock of waste bales.

16. Method for training machine learning classifiers (14) for use in a method according to any of the claims 8 to 15, characterized in that the machine learning classifiers are trained by one or more of the following:- training, preferably in a supervised manner, using labelled data, wherein the labelled data typically comprises bale instances labelled by a human expert, and / or- training, preferably in a supervised manner, using synthetically generated data, wherein the synthetically generated data typically comprises synthetically generated bale instances of waste bales or of waste, and / or- training in an unsupervised manner, in particular training on finding similarities in different images of one and the same bale and / or in different images of multiple bales, and / or- training, preferably in a supervised manner, using labelled data, wherein the labelled data typically comprises bale instances (24) combined with individually detected waste objects used to create, during a bale creation process, the bales to which the bale instances (24) correspond, and / or- training, preferably in a supervised manner, using labelled data, wherein the labelled data typically comprises bale instances (24) combined with information gathered during any of the processes and / or subprocesses and / or routines and / or steps of the method for analysing waste, in particular information gathered during the bale instance analysis process (6), or during a subsequent processing of a bale.

17. Image creation system (15), configured to create at least one image of at least one waste bale segment (23), wherein the system comprises a camera system (26), wherein the camera system (26) comprises at least one stationary camera and / or at least one moveable camera and / or at least one handheld camera and / or at least one vehicle-mounted camera.

18. Bale instance creation system (16), configured to receive as input an image, preferably an image created by the image creation system (15) according to claim 17, and to create a respective bale instance (24) for each waste bale segment (23) present in the image.

19. Bale instance analysis system (17), configured to receive as input a bale instance (24), preferably a bale instance (24) created by the bale instance creation system (16) according to claim 18, and to determine a bale parameter set for the bale instance (24).

20. Waste bale analysis system (18), comprising an image creation system (15) according to claim 17 and a bale instance creation system (16) according to claim 18 and a bale instance analysis system (17) according to claim 19 and preferably a database (19) configured to hold a multitude of bale parameter sets for a multitude of bale instances (24) and preferably a visualization system (20) for visualizing the database (19), at least partly, wherein the waste bale analysis system preferably comprises a recommendation system (22) comprising means configured to carry out a recommendation process during which, preferably based on the bale parameter set, a recommendation for a storage position inside a stock of waste bales for the bale to which the bale instance (24) corresponds is made, wherein the recommendation process preferably comprises a comparison between the bale parameter set and bale details, in particular bale parameter sets, of a multitude of waste bales in the stock of waste bales.

21. Computer program, comprising instructions which, when the program is executed by a computer, cause the computer to carry out a method according to any of the claims 1 to 16.

22. Computer-readable medium, comprising computer program code for carrying out a method according to any of the claims 1 to 16 and / or comprising a computer program according to claim 21 .

Images