WO2022112828A1

WO2022112828A1 - The technology of detecting the type of recyclable materials with sound processing

Info

Publication number: WO2022112828A1
Application number: PCT/IB2020/061148
Authority: WO
Inventors: Mohammad Bagher ESKANDARI
Original assignee: Eskandari Mohammad Bagher
Priority date: 2020-11-25
Filing date: 2020-11-25
Publication date: 2022-06-02
Also published as: GB202307140D0; GB2616534A

Abstract

This disclosure statement is about a developed technology applied in order to recognize the material types of the recycling objects by using the sound processing with two modes of item delivery including one at a time and in bulk in the reverse vending machine. In this technology, sound signals are generated by using the stresses imposed on the recycling materials with different procedures. After the signals are sensed by the main recognition system, it is possible to compare, and finally, recognize the material types of the recycling items and separate them. This system is developed by using the unique recognition pattern through the statistical methods.

Description

THE TECHNOLOGY OF DETECTING THE TYPE OF RECYCLABLE

MATERIALS WITH SOUND PROCESSING

TECHNICAL FIELD

[0001] This disclosure statement describes generally the technology for recognizing the materials of the recycling items. It explains specifically a technology for identifying the materials of the recycling objects by using the sound processing approach in the Reverse Vending Machines (RVM) that can accept the recycling items with two delivery modes including in bulk and one item at a time.

BACKGROUND

[0002] The Reverse Vending Machines (RVMs) have been operated for many years in the world and the applied aim is oriented to collect empty beverage bottles and containers. These machines can accept empty beverage bottle containers made of glass, plastics, and aluminum (beverage cans) as one item at a time (one piece) and reward some cash or give credit to the citizens after the kind of bottle is identified.

[0003] In the reverse vending machines, three technologies are commonly used for recognizing the materials of the recycling objects: The barcode recognition technology, the recognition technology based on image processing, and their integration, all of which are developed in order to detect the kinds of beverage bottles and containers.

[0004] The mentioned technologies have some technical and economic disadvantages as follows:

The technical disadvantages of these technologies: Firstly, their functions strongly depend on the physical container shape. In the other words, the material type of a deformed bottle cannot be identified because of the imposed compressive stresses. Secondly, it is possible to recognize the plastic and PET bottles and differentiate the plastic ones from PET ones only with barcode reading and if the barcodes cannot be read for some reasons, the material types of empty bottles cannot be identified. However, the PET bottles are much more valuable than the other plastic bottles. Thirdly, the existing reverse vending machines can only accept the bottle containers as one item at a time resulted in a time-consuming delivery in case of large quantities of these items.

[0005] Their economic disadvantages contribute to this fact that these machines just collect beverage bottles. Therefore, a wide range of the recycling objects that are mainly produced in the household has been removed from this mechanized cycle. In one of the general procedures, the processing function is performed on the sound signals that are reflected from the recycling items in order to recognize their material types as described below. First, some procedures should be chosen to produce the sounds that can be adjusted with the physical behaviors and specifications of the recycling materials as extracted features from the sound signals can be differentiated from each other as much as possible. Then by developing an appropriate recognition pattern which is applied integrated with the previously mentioned procedures, a high-efficient recognition system can be achieved in order to identify the material types of all recycling objects and resolve the previous disadvantages.

SUMMARY

[0006] In this disclosure statement, a new sound processing-based technology is provided to recognize the material types of the recycling items based on two general modes: one mode for the recognition of the recycling items delivered in bulk (multiple of the same items collected in a bag), and another mode for the recognition of the recycling items delivered as one item at a time (one piece). In fact, the proposed technology can operate with these two delivery modes without any required adjustments in the process of the recognition system. [0007] In this technology, different procedures were studied in order to produce the sound signals and extract their features. Finally, two integrated methods of sound production were selected that were completely different and useful regarding the produced features. Therefore, a distinctive recognition pattern which was able to extract features was provided. It should be noted that the mentioned recognition pattern was determined through accurately studying the behaviors of recycling materials against the compressive stress by using the statistical methodologies.

[0008] Generally, this technology intends to recognize five main classes of the recycling materials including paper and cardboard, PET, glass, plastics, and metal in delivery mode as in bulk, and also all kinds of bottles made of glass, plastics, aluminum, and PET and the coffee cups in delivery mode as one item at a time. The two -mode recognition procedure was performed in the reverse vending machine with over 97% recognition efficiency.

BRIEF DESCRIPTION OF THE DRAWINGS

[0009] FIG. 1 illustrates hardware block diagram.

[00010] FIG. 2 illustrates main recognition system block diagram.

[00011] FIG. 3 illustrates MFCC algorithm block diagram.

[00012] FIG. 4 illustrates feature extraction block diagram.

[00013] FIG. 5 shows LBG method.

[00014] FIG. 6 is flow chart of falling voice recognition pattern.

[00015] FIG. 7 is flow chart of pressing voice recognition pattern.

[00016] FIG. 8 is flow chart of decision strategy.

[00017] FIG. 9 is table representation of executive test result.

[00018] FIG. 10 illustrates a schematic representation of mechanical process of technology (bulk model).

[00019] FIG. 11 illustrates a schematic representation of mechanical process of technology (one item at a time model). DETAILED DESCRIPTION

[0010] The processing of audio signals that reflect from the recycling materials can be considered as one of the common methods to recognize the materials of the recycling items. This method does not depend on the physical characteristics of the objects and the deformed physical shape cannot affect negatively on the recognition results of the recycling material types. On the other hand, it is possible to recognize rapidly the material types of the recycling materials delivered in bulk. In addition, the materials like PET and plastics can be differentiated accurately from each other. Accordingly, it can be claimed that the effectiveness of the recognition system based on the sound processing is higher than the existing technologies due to these capabilities. However, there are some main problems to develop this recognition system described as follows. There are many kinds of sound generation processes as following examples: Applying a pneumatic jackhammer to knock down the object, vibrating the object, sprinkling the water or sand onto the object, pressing the object, knocking down the object by using gravity force and so on. However, it is very important and complicated to select and apply the mentioned methods in the reverse vending machines due to the technical and economic limitations and also the technical efficiency.

[0011] The selection of one or more methods to generate audio signals (even those methods used in this technology) cannot achieve a high efficiency without the appropriate recognition pattern. It can be explained with high overlaps that are observed in the extracted features of the sounds that are reflected from different objects by using most sound generation procedures; consequently, these observed overlaps prevent a high efficient recognition system being developed for all kinds of recycling materials easily and merely by using one or more sound generation methods. Thus, the sound generation methods have to be selected in a way that the extracted features from the sounds can be differentiated from each other as much as possible. Then, the observed overlaps can be removed by using the appropriate recognition pattern according to the behaviors and specifications of the recycling materials, and a high efficiency can be achieved with the identification of all recycling objects. We could realize this aim and developed the mentioned technology in a 5-year research program.

[0012] The proposed technology consists of the main recognition system with three basic sections in order to recognize the material types of the recycling objects with two delivery modes as in bulk FIG. 10(A) or as one item FIG. 11(A) at a time as shown in the block diagram of FIG. 2. These sections include the sound generation 21,24. the sound feature extraction 22,25, and the development of the appropriate pattern in order to use the extracted sound features. The later section consists of three parts including the recognition pattern of falling step 23, the recognition pattern of pressing step 26, and the decision strategy 27. [0013] However, in another division, the main recognition system of this technology FIG. 2 can be divided into three sections as follows: the recognition system of falling step 21,22,23, the recognition system of pressing step 24,25,26, and the decision strategy 27. Sound generation

[0014] As explained before, different procedures were experimented to generate sounds in this technology, but the mechanism of falling of the recycling objects from a suitable height FIG. 10(B), FIG. 11(B) and the mechanism of pressing of the recycling objects FIG. 10(D),

FIG. 11(D) indicated the best efficiency values among the experimented procedures. In the mechanism of falling of the recycling items FIG. 10(B), FIG. 11(B) from a suitable height (nearly 50 cm), the stress is imposed on the recycling materials (delivered as one item at a time or in bulk) and the sound signals reflect with an effective time of nearly 1.5 seconds as an impulse train. The signals have short periods and high amplitudes in this impulse train. [0015] In the mechanism of pressing of recycling materials FIG. 10(D), FIG. 11(D), the stress is imposed on the recycling objects (delivered as one item at a time or in bulk), and their bulk volumes are reduced with some reflected sounds from them. This procedure achieved the sounds generated with lower amplitudes and higher periods compared to the generated sounds by using the previous method, and the generated sounds are approximately 6-8 seconds in duration that is much longer effective time compared to the received signals in the falling step.

[0016] According to the block diagram shown in FIG. 1, the sound signals that are reflected from the recycling materials in both mentioned mechanisms FIG. 10(B), FIG. 11(B), FIG. 10(D), FIG. 11(D) are recorded by using an advanced audio system including a digital microphone 34 and a digital sound card 33 with adjustable sensitivity. Generally, the sounds generated in the mentioned mechanisms can be used in both training and recognition sections. [0017] In training section, the related recycling materials are required to be passed through both mechanisms of falling and pressing steps FIG. 10(B), FIG. 11(B) , FIG. 10(D), FIG. 11(D), delivered as one item FIG. 11(A) at a time or in bulk FIG. 10(A), and then, the appropriate sounds generated in this process should be recorded and saved by using an industrial computer 32. This performance is required to be repeated many times with acceptable rounds, and about 1500 sound samples are generated for each recycling objects in this technology. Then, these generated sounds are used to train the main recognition system FIG. 2 in order to identify the recycling materials 28. Training the main recognition system FIG. 2 should be continued in order to identify the recycling materials 28, until the system shows the appropriate efficiency; therefore, when this aim is achieved, there is no need to continue re-train and make more changes.

[0018] In the recognition section, the recycling objects are required to be passed through both mechanisms FIG. 10(B), FIG. 11(B), FIG. 10(D), FIG. 11(D) (falling and pressing), delivered as one item FIG. 11(A) at a time or in bulk FIG. 10(A), and subsequently the audio signals generated from this process should be analyzed by using the recognition system FIG. 2 to identify the type of material 28. In addition, all actuators 31 controlled by a main board 29 and powered by proper driver boards 30 lead to creating appropriate actions according to the recognition process. Sound Feature Extraction

[0019] The sounds recorded in the falling and pressing steps 14 are entered into a function in order to extract the sound features with the following steps:

In the first step of the feature extraction, Mel Frequency Cepstral Coefficient (MFCC) is extracted. The MFCC is designed based on human auditory features in the speech recognition and understanding.

[0020] According to the block diagram in FIG. 3, in the first step of the proposed algorithm, the recorded audio signals 14 are divided into some different frames 15. Then, hamming window function 16 is used to reduce the effect of discontinuity at the edges of each frame. Next, the FFT 17 of signal spectrum is gotten. Consequently, the signal spectrum is passed through the Mel filter 18 and finally, the DCT 20 is computed on the spectrum.

[0021] As many features are extracted in the MFCC step, it is required to select and classify the appropriate and differentiated features. Therefore, the LBG ¹ method is a suitable procedure for this purpose and the method function can be observed in FIG. 5.

The LBG output 9 constitutes the feature vector 13 integrated with standard deviation 10, and minimum/maximum MFCC outputs 12,11. This vector shows the output of the feature extraction process according to the block diagram in FIG. 4.

[0022] Now, we can compare the obtained feature vector 13 with the previously trained samples in order to extract various possible material types with their corresponding rates of the recognition accuracy.

[0023] The mentioned function should be performed on the sounds generated in both falling and pressing steps separately. According to some reasons detailed later, the mentioned function does not achieve the appropriate results for all kinds of the recycling objects without the appropriate recognition pattern 23,26. In other words, this function alone cannot result in a high-efficient recognition system for the recycling items.

Development of the appropriate pattern for using the extracted sound features

[0024] This section includes the recognition pattern of falling step 23 , the recognition pattern of pressing step 26 , and the decision strategy 27 that in follows will be described.

[0025] Now, we have two different inputs from the extracted features of the sounds generated in the falling and pressing steps. First, both inputs are analyzed as follows:

Input analysis in the feature extraction of the sounds generated in the falling step [0026] The extracted data in this step show a major drawback being necessary to resolve. If not, they cannot be used about three material classes of metal, glass and the illegal objects

¹ Linde, Buzo & Gray [0001]

(stones and bricks), because their recognition efficiency values become very low. The problem is related to couple of existing overlaps in the extracted features of the sounds that are generated from these recycling materials in the falling step. This issue can rise to the false recognition.

[0027] This overlap is observed only in the recycling materials with harder material types because the remarkable stress is imposed on the storage chamber 3 that receives the fallen recycling materials. As the most industrial machines are made of metal, the sound generated from the imposed stress on the chamber 3 takes a longer fraction from total useful time of 1 second. In this state, the result from the recognition system of falling step is proved to be the metal material with a remarkable accuracy, because there is a high similarity between the mentioned sounds with the trained metal sound samples. If there are high density and rigidity values for those points of the recycling objects stricken on the chamber 3 after falling, the sounds are already reflected from the chamber 3 instead of the recycling objects.

[0028] To overcome this problem, a metal sensor 6 was used as an auxiliary sensor in order to remove these overlaps. Of note, if we change the material of the chamber 3, the stressinduced sounds will be reflected from the chamber 3, again. But the difference is that a sensor like the metal sensor 6 which is corresponding to the type of chamber is not existed to be installed to help the differentiation of the observed overlaps. If some procedures are performed and the chamber 3 functions as a bumper to reduce the stress from falling of the recycling objects, the sound generation mechanism cannot be reliable anymore and obtains very low recognition efficiency.

[0029] A metal sensor 6 as an auxiliary sensor can be applied with an integrated appropriate procedure in order to categorize and prioritize the material classes; consequently, the problem is removed and the efficiency of the recognition system is increased in order to separate the recycling materials in falling step. We named above integrated procedures as the recognition pattern of falling step FIG. 6.

The recognition pattern of falling step

[0030] This recognition pattern FIG. 6 is performed by using two procedures, including the optimization procedure of the number of existing classes which is used the auxiliary sensor 6, and an efficient procedure of categorization and prioritization of different classes.

Optimization of the existing class numbers with the auxiliary sensor

[0031] In this step, the initial recognition is performed to find out whether or not, there are metal objects in the recycling bag (in recognition with the bulked items FIG. 10(A)), or the beverage bottles FIG. 11(A) (in recognition with one item at a time) at the start of the recognition process. Then, the number of classes can be optimized and the sounds generated in the falling step can be analyzed with the higher accuracy in the remained classes. Of note, it is not enough to apply only the metal sensor 6 for recognizing whether or not the recycling objects are made of metal, because the other recycling items sometimes contain some small metal parts; For example, notebooks with spiral metal binding, plastic containers with aluminum coating, the glass bottles with metal caps, and so on. Therefore, the audio signal analysis is required to perform integrated with this sensor in order to identify the major materials of the recycling objects, so the sensor 6 can be called the auxiliary sensor.

An efficient procedure for categorization and prioritization of the different classes [0032] Based on the statistical studies performed on the behaviors of the recycling materials in the falling step, an appropriate procedure is defined for the categorization and prioritization of the existing classes which the main criterion of the classification and prioritization of the classes is "Separability parameter".

Description of the recognition pattern of falling step

[0033] The recognition pattern of falling step for the recycling materials can be observed in a flowchart represented in FIG. 6. According to FIG. 10(C), FIG. 11(C) the procedure of the recognition pattern is described as follows: first, the metal sensors 6 attach to the recycling materials which is used the suitable mechanical mechanism, then they can identify whether or not, the metal material is used in the recycling objects.

[0034] If the metal material is not identified in the recycling objects, the existing classes are categorized in to four classes of "Plastics", "Glass", "Paper & Cardboard", and the "Illegal Objects". Based on this condition, the recognition system of falling step determines the final recognition results of the recycling materials according to the results obtained from analyzing this group.

[0035] In this class categorization, two classes of "PET" and "Metal" are removed according to the following reasons:

[0036] The class of "Metal" is removed from this classification because of the overlap it showed with the classes of "Glass" and the "Illegal objects" (stones and bricks) regarding the extracted features of the generated sound in the falling step. As the metal materials are not identified in the recycling materials by the metal sensor 6, the removal of "Metal" from the class categorization is made with a high confidence.

[0037] Also, the class of "PET" showed a high overlap with the class of "Plastics" regarding the extracted sound features in the falling step because of their high essential similarities observed in the constituted ingredients. As the "Separability parameter" of the class of "Plastics" seems to be better than that for the class of "PET" regarding the extracted sound features in the falling step, the class of "PET" was removed from all categorizations performed on the recycling materials in the recognition pattern of falling step.

[0038] However, if the metal sensor 6 verifies the presence of metal materials in the recycling objects, it is required to determine whether or not, the metal is the major constituent or only a small part in the recycling materials according to the previous description.

[0039] Therefore, we first restricted the class categorization into three classes of "Metal",

"Plastics", and "Paper & Cardboard" with the assigned higher priority to this group. Therefore, if the classes of "Paper & Cardboard" or "Plastics" are outputted in the recognition result of the material types in this group, the final result will be the same in the falling step, but if the class of "Metal" is outputted as the recognition result, it should be verified in the second group that includes the class of "Glass", too. In this state, the second group with the lower priority determines the final recognition result of the recycling materials in the falling step.

Input analysis of the extracted features from the sounds generated in the pressing step

[0040] The extracted data in this step showed some major drawbacks like ones in the falling step as follows: The generated sounds from the recycling materials of glass, illegal objects (stones and bricks) and metal cannot be used in the pressing step, at all.

[0041] The mentioned recycling materials have high strength; therefore, when they are pressed, the distinctive sounds cannot be generated and also the glass cannot be imposed under a destructive stress. In other words, the extracted features from the sounds that are reflected in these classes show the intensively high overlaps with each other, and to some extent, with the other recycling materials. This issue may result in the highly reduced efficiency of the recognition system of pressing step.

[0042] Based on the statistical studies performed on the behaviors of the recycling materials in the pressing step, an appropriate procedure is defined for the classification and prioritization of the existing classes and the "Separability parameter" is the main criterion for the classification and prioritization of the classes. This is called the recognition pattern of pressing step FIG. 7 for the recycling materials.

The recognition pattern of pressing step for the recycling materials

[0043] The recognition pattern of pressing step for the recycling material can be observed with a different priority in two groups in a flowchart represented in FIG. 7 as follows:

[0044] First, the classes are categorized into two classes of "PET" and "Plastics" having a higher priority assigned to this group. Thus, if the recognition output includes "PET" in this group, the final recognition result is the same in the pressing step. However, if the recognition output includes "Plastics", this result should be proved in the second group including the class of "Paper & Cardboard", too. In this state, the second group with the lower priority determines the final recognition result of the recycling items in the pressing step.

[0045] As the class of "PET" has the highest "Separability" in the pressing step and shows the lowest "Separability" in the falling step among all recycling classes, it belongs to the first group with the highest prioritization in this recognition pattern.

[0046] At this point, we studied precisely two inputs obtained from the feature extraction in the falling and pressing steps. Also, the related drawbacks and their solutions were presented, and based on the provided solutions, we achieved two recognition systems of falling step and pressing steps having the complete independence in the identification of the recycling materials including some information like the recognized material types and their recognition percentages.

[0047] Because there are two different and independent recognition systems developed for the falling and pressing steps FIG. 2, it is clearly necessary to determine the threshold and prioritization about the decisions made by these systems in order to finally determine the recycling material type. In other words, when the threshold is determined, and the decisions are prioritized in these recognition systems, the valuation function has already been performed about these decisions. This valuation process is based on the precise statistical analysis performed on their functions for the recognition of the existing classes. This process is considered to be a "decision strategy".

Decision strategy

[0048] The Decision strategy is prioritized according to the flowcharts represented in FIG. 8 in seven steps as follows: First, the recognition system of falling step is put in top prioritization because of the higher accuracy, and also the recognition of those classes being forbidden to enter to the pressing step.

[0049] The classes like "Glass" and the "Illegal objects" (stones and bricks) are not permitted to be pressed, at all. Because the colored and non-colored recycling types of glass become useless when are pressed and mixed, due to different melting temperatures. On the other hand, it is useless to press the stones and bricks, because it results in some serious damages to the pressing system installed in the RVMs.

[0050] According to the flowchart represented in FIG. 8, seven states can result in the determination of the outputs in this decision strategy and the possible outputs include glass, the illegal objects (stones and bricks), paper and cardboard, PET, metal, plastics, and others. [0051] Six mentioned outputs are defined for the system and the class of "Others" will be outputted by the recognition system when this system is not able to determine the recycling material type according to the defined conditions in the Decision strategy. In this state, the bag or object delivered to the RVM should be transported to the container dedicated to "Others" that can be shared with the container for the "Illegal Objects".

[0052] In addition to the prioritization, the thresholding has a remarkable effect on the final results in the Decision Strategy Section. Because there are different accuracy values for the recognition outputs of the recycling objects in two recognition systems (the falling or pressing steps) regarding different existing classes and the reliability values of these decisions influence remarkably on the recognition efficiencies regarding the materials of the recycling objects. It can be realized by determining the constant threshold (or the recognition accuracy) for each class in both recognition systems of falling or pressing steps.

[0053] Of note, the smaller constant threshold mean the higher reliability in the recognition results obtained from the related recognition system (the falling or pressing steps).

[0054] For example, the constant threshold is set as 15 for the material type of "Glass" in the first priority of the Decision Strategy indicated the high reliability and accuracy of outputs that the recognition system of falling step showed for the material type of "Glass".

[0055] Now, the flowchart represented in FIG. 8 is described as follows:

[0056] If the material type of "Glass" is recognized at A²>15% in the falling step, the final recognition result will be the same.

² Accuracy [0057] If the material type of the "Illegal objects" is recognized at A>55% in the falling step, the final recognition result will be the same.

[0058] If the material type of "Paper & Cardboard" is recognized at A>75% in the falling step, and also the material type of "Paper & Cardboard" is recognized at A>50% in the pressing step, the final recognition result will be the material type of " Paper & Cardboard" by the main recognition system.

[0059] If the material type of "PET" is recognized at A>55% in the pressing step, the final recognition result will be the same.

[0060] If the material type of "Metal" is recognized at A>75% in the falling step, and also the inductive sensor can recognize the material type of "Metal", the final recognition result will be the material type of "Metal" by the system.

[0061] If the material type of "Plastics" is recognized at A>55% in the falling or pressing steps, the final recognition result will be the same.

[0062] And finally, if none of the mentioned cases occurred, the main recognition system is actually not able to determine the material types of the recycling items; therefore, the option "Others" is outputted by the system.

RVM function and the performance test on the recognition technology for the recycle materials delivered as in bulk

[0001] path. Consequently, the stress is imposed on the recycle items and makes the sounds to be generated in the falling step. The generated sound is recorded and analyzed by using the recognition system of falling step. Then, the horizontal pneumatic jackhammer 4 encircles the bag, and if the recognition result is not outputted as the material types of "Glass" or "Illegal objects" by the recognition system of falling step, the pressing function is performed on the recycling materials in the bag by using the vertical pneumatic jackhammer 5. In this case, when the stress is imposed on the recycling items, it generates the sound related to the pressing step and the generated sound is recorded and analyzed by using the recognition system of pressing step. Then, both vertical and horizontal jackhammers are returned back to their initial places. [0064] Now, the recognition system of pressing step represents its recognition results, and according to FIG. 2, the results obtained from both recognition systems of falling and pressing steps are entered to the Decision Strategy Section in order to determine the material types of the recycling items by the main recognition system.

[0065] Of note, the recognition process for the recycling materials is similar for both delivery modes of one item at a time and in bulk and includes five steps according to FIG. 11 (A-E). [0066] The function results of this technology in the reverse vending machine can be observed in the table shown in FIG. 9. As shown in this table, many samples were prepared with remarkable numbers as nearly 3000 for the implementation test in each class and were put in the bags with specific dimensions (200 bags for each class). Then, the bags were delivered to the RVM machine for 250 times in each class, and the recognition efficiency of this technology was measured for each recycling material type as 97.12%, 99.06%, 100%, 100%, 99.52%, and 99.51% in the classes of "PET", "Plastics", "Paper & Cardboard" (Mixed or Separability), "Metal", "Glass", and the "Illegal Objects" (stones and bricks), respectively.

Claims

1- The technology is based on the sound processing and is used for the recognition of the recycling materials with two delivery modes including one at a time (one item) and in bulk (multiple of same items in a bag). This technology includes three sections as follows: the sound generation, the feature extraction from the sounds, and the appropriate pattern applied in order to use the extracted features from the sounds. The mentioned technology can be applied in the reverse vending machines (RVM) with two delivery models including one at a time and in bulk. When the delivery mode is adjusted as one item at a time, the classes of glass, PET, and plastic bottles, as well as aluminum cans, coffee cups and illegal objects (stones and bricks) can be recognized. When the delivery mode is adjusted as in bulk, the classes of glass, plastics, metal, paper and cardboard, the illegal objects (stones and bricks), and PET can be recognized.

2 According to Claim 1, the sound generator includes two different mechanical mechanisms that can generate the sounds with different features by imposing stress on the recycling materials.

These mechanical mechanisms include the mechanism of falling of the recycling materials from an appropriate height (nearly 50 cm) and the mechanism of pressing the recycling materials.

3 The method of Claim 1, the sound features are extracted as follows: the sounds obtained from the stresses imposed on the recycling materials (see Claim 2 for the detailed description) will enter into function Separability with the output as the feature vector.

In order to obtain this feature vector, first, the input signals are undergone the MFCC, then those extracted sound features that are proved to be appropriate and differentiated, are selected and categorized by using the LBG method. Finally, the LBG output is used with the standard deviation, and the minimum/maximum MFCC outputs to form the feature vector.

4 According to Claim 1, in order to use the extracted sound features (see Claim 3 for the extraction methodology), the suitable pattern has three sections including the recognition pattern of falling step, the recognition pattern of pressing step, and the decision strategy. The recognition patterns of falling and pressing steps can be used as the solutions for the recovery of overlaps seen in the sound features of the recycling materials. Finally, the decision strategy can be used in order to recognize the recycling materials with the high efficiency in two delivery modes including one at a time and in bulk.

5 According to Claim 1, the recognition pattern of falling step includes two main parts including the optimization of the number of existing classes by using the auxiliary sensor, and then an efficient procedure for the classification and prioritization of the different classes found by using the statistical studies on the behaviors of the recycling materials. The auxiliary sensor is a metal sensor that can recognize whether or not, there are metal objects in the recycling bag (with the recognition of bulk items), or in the beverage bottle (with the recognition of items accepted as one at a time) at the start of the recognition process. Subsequently, the sensor makes that the number of classes can be optimized and the sounds generated in the falling step can be analyzed with the higher accuracy in the remained classes.

The different material classes are categorized as follows: If any metal is not identified in the recycling objects, the existing classes are classified into four classes of "Plastics", "Glass", "Paper & Cardboard", and "Illegal Objects". Based on this condition, the recognition system of falling step determines the final recognition results of the recycling materials according to the results obtained from analyzing this group.

However, if the metal sensor verifies the presence of metal in the recycling materials, it is required to determine whether or not, the metal is the major constituent or only a small part in the recycling materials.

Therefore, first we restricted the categorization of classes into three classes of "Metal", "Plastics", and "Paper & Cardboard" with the assigned higher priority to this group. Therefore, if "Paper & Cardboard" or "Plastics" are outputted as the recognition results of the material types in this group, the final result will be the same in the falling step, but if the "Metal" is outputted as the recognition result, this output should be verified in the second group that includes the class of "Glass". In this case, the second group with the lower priority determines the final recognition result of the recycling materials in the falling step.

6- According to Claim 1, the recognition pattern of pressing step was performed in two groups with different prioritizations based on the statistical studies on the behaviors of the recycling materials in the pressing step.

The first group includes two classes of "PET" and "Plastics" with the higher priority, the second group includes two classes of "Plastics" and "Paper & Cardboard" with the lower priority. These groups determine the final recognition results of the recycling material types in the pressing step.

7- According to Claim 1, the decision strategy includes two sections which are thresholding and prioritization.

Since these recognition systems work independently and their efficiencies are different for existing material classes; the threshold value was determined and the decisions were prioritized in both recognition systems of falling and pressing steps and in fact the valuation function has been performed on these decisions. The decision strategy includes 7 steps as follows:

If the material type of "Glass" is recognized at A>15% in the falling step, the final recognition result will be the same.

If the material type of "Illegal objects" is recognized at A>55% in the falling step, the final recognition result will be the same.

If the material type of "Paper & Cardboard" is recognized at A>75% in the falling step, and also the material type of "Paper & Cardboard" is recognized at A>50% in the pressing step, the final recognition result will be the material type of "Paper & Cardboard" by the main recognition system.

If the material type of "PET" is recognized at A>55% in the pressing step, the final recognition result will be the same. If the material type of "Metal" is recognized at A>75% in the falling step, and also the inductive sensor can recognize the material type of metal, the final recognition result will be the material type of "Metal" by the main recognition system.

If the material type of "Plastics" is recognized at A>55% in the falling or pressing steps, the final recognition result will be the same.

And finally, if none of the mentioned cases occurred, the main recognition system actually is not able to determine the material types of the recycling items; therefore, the option "Others" is outputted by the system.