Intelligent Kitchen Waste Composting System via Deep Learning and IoT

doi:10.21203/rs.3.rs-2978777/v1

Download PDF

Research Article

Intelligent Kitchen Waste Composting System via Deep Learning and IoT

https://doi.org/10.21203/rs.3.rs-2978777/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 07 Dec, 2023

Read the published version in Waste and Biomass Valorization →

You are reading this latest preprint version

Kitchen waste is listed among the top global sustainability issue as it contributes to global warming and climate change. Composting is one of the solutions to tackle the issue of kitchen waste increment. However, a manual composting system has led to several problems for the waste management authorities to invest more in human labor, cost, and time to segregate and dispose of the kitchen waste and its composting soil. Therefore, this project proposes an intelligent kitchen waste composting system via deep learning and Internet-of-Things (IoT) that is fully automated to cater for that issue. Firstly, the proposed system utilized Convolutional Neural Network (CNN) to detect and segregate kitchen waste into compostable and non-compostable categories. Then, the classified compostable waste went through composting stage inside an automated compost bin with the feature of IoT. The IoT compost bin requires less human labor as it used sensors, actuators, and Wi-Fi connection to monitor and control the composting process. Finally, the compost soil is transferred to the designated gardening area via smart compost soil transportation system. The system consists of a robot equipped with infrared sensors. The sensors control the robot's movement by tracking the predefined black tape path. A prototype is built to investigate the performance of the proposed system. Results show that each sub-system managed to interact with one another, thus creating a large intelligent system that succeeded in completing the kitchen waste segregation, composting and ready compost delivering tasks automatically. In the future, it is expected that the proposed intelligent system has the potential to be commercialized to tackle the kitchen waste increment issue as it offers an economical yet high-efficiency solution.

Kitchen Waste

Composting

Convolutional Neural Network

Internet-of-Things

Automation

Application of Deep Learning, Convolutional Neural Network to automatically segregate kitchen waste.
Implementation of Internet-of-Things (IoT) to remotely monitor composting process.
Mobile robot to complete ready compost transportation system

Solution for traditional composting process where deep learning is implemented to segregate waste, Internet-of-Things used to monitor composting and robot developed to transport the ready compost.

The food waste (FW) trend in Malaysia advanced to 16,667.5 tonnes daily in 2020 and it was reported that the household sector accounts for 44.5% of it (Bernama 2020). The increment and mismanagement of the waste are harmful to the environment because it produces a greenhouse gas called Methane which can contribute to global warming and climate change (Secretariat 2022).

Malaysia government together with the non-governmental organizations initiated multiple composting projects in zeroing the FW issue. In 2016, Malaysia government collaborated with Japan to launch a project which promotes composting as a zero FW solution in Fraser Hills (Nakanishi 2017). In 2021, Menteri Alam EKO Sdn Bhd (MAEKO) is one of the Malaysia's companies that successfully invented composting technology machines that can convert organic waste into compost (MAEKO 2022).

Based on the above matter, it can be concluded that composting can be one of the solutions for the FW issue. However, a manual composting system creates a burden to the authorities as they need to invest in human labor, cost, and time. Thus, this study proposes an intelligent kitchen waste (KW) composting system incorporated with deep learning and Internet-of-Things (IoT) technologies. KW is targeted as it has wide variation compared to the FW. KW is described as the leftover matter that has been thrown from the kitchen before the consumer serving process. The waste is normally produced by households, hotels, restaurants, and every area that involve the cooking process (Li et al. 2009).

The proposed intelligent system is compromised with three subsystems: kitchen waste segregation, composting, and ready compost transporting. Firstly, the kitchen waste segregation system utilized a deep learning method called Convolutional Neural Network (CNN). CNN is chosen because it is widely used and had shown impressive performance in detection and classification (Yao et al. 2019; Novianti et al. 2015).

The second subsystem consists of an IoT compost bin to perform the KW composting process. Composting is the natural process of leveraging microorganisms to decompose organic matter into compost, which is the soil's nutrient rich fertilizer (Ayilara et al. 2020). IoT can be defined as a system of interconnected embedded devices that exchanges data over a network through communication protocols (Svendsen et al. 2021). The incorporation of an IoT system can reduce human effort significantly as it allows users to monitor the real-time sensors readings and control the actuator remotely via a web application (Gubbi et al. 2013).

The third subsystem is the compost soil transportation system. The rapid progression in IoT and robotics technologies led to smarter agricultural equipment inventions (Kumar and Ashok 2021). This project developed a line-following robot equipped with an infrared (IR) sensor to transport composting soil to the prescribed gardening location. The IR sensor is used to sense and track the pre-defined path. Track-following robot equipped with an IR sensor has been used in several applications, such as goods transporter (Hanumante et al. 2013), services robot, household chores, healthcare management system (Punetha et al. 2013), and agricultural robotics system (Adabara et al. 2019). This kind of robot can increase work efficiency, saves time, and reduce human labor (Abraham et al. 2020).

Overview of the system

Figure 1 displays the prototype of the intelligent kitchen waste composting system. The prototype can be categorized into three subsystems. The red box indicates the Kitchen Waste Segregation System, whilst the blue box represents the composting system, and the green box represents the compost soil transportation system. The description for each system is listed in Table 1. Figure 2 shows the overall system's fundamental operation. The CNN Model was trained with TensorFlow library in Jupyter Notebook platform using transfer learning method, the CNN model used is MobileNetV2. The total number of images used in training and validating the model were 985, 68% for training and 32% for validation. These images were collected and labelled locally to classify the kitchen wastes into two categories which are compostable such as banana, carrot, cucumber, eggshell, lemon, lime, orange, and potato, whereas non-compostable such as food cans, sauce bottles, food packaging plastic and spice bottle. The system begins with an ultrasonic sensor to detect the presence of the kitchen waste and trigger the object detection element. The deep learning algorithm of the system will classify the kitchen waste as compostable or non-compostable. If the kitchen waste is not compostable, it will be segregated into the dumping bin. Otherwise, the compostable kitchen waste will be shredded to a smaller size of waste, which will ease and speed up the composting process. Before commencing the composting process, the server communication will link the preceding system with the composting process and compost soil transporting process through IoT integration. The composting process will send real-time sensor data to the server. When the compost is finished, the server will send a signal to the compost soil transporting process to collect the compost soil and deliver it to the designated location.

Table 1

Prototype system description
No.	Systems Name	Description
1	Kitchen Waste Segregation System	AI and IoT technology have been integrated into the system. The trained deep learning model can classify kitchen waste into compostable and non-compostable with camera vision, to segregate the kitchen waste. A signal will send to the server once completed the instruction of this system.
2	Composting System	The composting system is linked to the kitchen waste segregation system. The composting system will process the kitchen waste into compost. Through the integration of AI and IoT technologies, the system monitors the compost state and may inform the user of the product's condition.
3	Compost Soil Transportation System	Following the production of compost, the robot transporter, which is linked to the composting system through an IoT system, will arrive and collect the compost soil. Following the black tape trail, the compost soil will be delivered to the destination.

The hardware of each sub-systems is stated in the sub-section as shown in Table 2 to Table 4:

Table 2

Kitchen waste segregation system components
No.	Component	Specification	Description
1	Raspberry Pi 4	• Quad core Cortex-A72 64-bit	Microcontroller
2	L298N Bridge Motor Driver	• Operating DC Voltage range: 5–35 V. • Maximum power 25 W.	Control stepper motor speed and direction.
3	Web Camera	• Resolution 1920 p ×1080 p. • Frame rate, 30 FPS	Run real-time video stream for object detection.
4	HC-SR04 Ultrasonic Sensor	• Operating DC voltage: 5 V • Detection range: 2 cm to 400 cm	To trigger the camera to run object detection model.
5	17HS4401 Stepper Motor	• Bipolar stepper • Holding torque: 0.4 Nm.	Act as the rotational actuator to turn the segregation platform.
6	Aluminum Profiles	• Size: 30 mm × 30 mm × 800 mm.	body frame’s support.
7	Acrylic Sheet	• Size: 420 mm × 594 mm (A2) • Thickness: 3 mm	For prototyping body frame.
8	Shredder Blades	• Diameter: 80 mm • Thickness: 10 mm	For shredding purposes.

Table 3

Composting system components
No.	Component	Specification	Description
1	ESP32	• 32-bit dual-core	Microcontroller
2	DHT22 Temperature and Humidity Sensor	• 3.3-6 V • Humidity from 0-100% Relative Humidity • Temperature from − 40°C to 80°C.	To measure the ambient humidity and temperature of the compost bin.
3	Capacitive Moisture Sensor	• 3.3-5 V • Analog I/O 0-4095.	To measure the moisture level of the compost pile.
4	DS18B20 Temperature Probe	• 3-5.5 V • Temperature from − 55°C to 125°C.	To measure compost pile temperature.
5	Brushed DC Motor	• 12V • 5.5 kgfcm	To rotate the compost bin.

Table 4

Compost soil transportation system components
No.	Component	Specification	Description
1	ESP32	• Xtensa dual-core 32-bit LX6	Microcontroller
2	TCRT5000 IR Sensor	• Peak Operating Distance: 2.5 mm. • Operating voltage: 5 V.	Using an infrared (IR) beam to detect the black tape path.
3	L298N Bridge Motor Driver	• Integrated 5V power regulator • 2A max drive current	Motor controller to regulate motor speed and direction.
4	TT Gear Motor	• 3–6V DC. • Operating Speed: 200 RPM.	Gear DC motor to move the transporter.

Software

Web server framework

Django was chosen as the framework for the web server. Django is a backend framework that is completely based on the Python programming language, making it very simple to build. In addition, Django has native support for SQL database management systems, making database interaction incredibly straightforward and requiring no SQL code to be developed.

Database management systems (DBMS)

DBMS is a system used to improve and manage database storage. MySQL DBMS was chosen because of the developer's expertise.

Communication protocol selection

The HTTP and MQTT communication protocols were selected as the major client-server and endpoint protocols, respectively. HTTP is a document-centric communication protocol, making it a sensible choice for the user side given that its main purpose is to supply content, in this case HTML pages. MQTT is a data-centric protocol that is more efficient than HTTP for sending data that consists just of data, such as JavaScript Object Notation (JSON).

System data retrieving technique selection

Short polling was selected as the data retrieving method to create the system rapidly due to the short time frame of the project.

TensorFlow

TensorFlow was selected as the library to perform the deep learning computation. TensorFlow is a sophisticated and highly effective deep learning package with robust visualization features and several alternatives for high-level model construction.

Kitchen waste segregation system

According to the results of validation and training that have been done, the trained model using the CNN architecture of MobileNetV2 achieved a mean average of total loss of 0.091 as shown in Fig. 3. In Fig. 4., the learning rate experienced a steady state condition from 1500 epoch to 4000 epoch with the learning rate of 0.079. Furthermore, Table 5 shown the evaluation matric for the trained model, it has successfully produced high score of 0.8279 and 0.8548 for mean average precision and recall respectively. Besides, to obtain an average harmonic between the precision and recall, the F1 score was calculated. Thus, the trained model obtained a great result of 0.8411 for the F1 score.

Table 5

The evaluation metrics of CNN model
CNN architecture	Epoch	Mean Average Precision	Mean Average Recall	F1 Score
MobileNetV2	4000	0.8279	0.8548	0.8411

After the training and validation process, the trained model was tested with orange and food packaging plastic. Figure 5 shows the trained model was able to classify and detect the orange as compostable, and food packaging plastic as non-compostable accurately with a higher confident level of 97% and 93% respectively.

Composting system

Figure 6. shows the compost bin web application. The data visualization part displaying the real time sensor readings asynchronously is located at the top center part of the web page whereas the actuator control switches are located at the mid center part of the web page. 2 dimensional and 3 dimensional graphs were also added at the bottom center part of the web page for visualization purposes. The green start button located at the bottom of the web application is the button to trigger the Compost Soil Transportation System.

A compost pile was initiated to observe the composting parameters trend for 7 days. Figure 7 shows the area graph plot of the data collected throughout the 7 days of composting period. Observing the Aeration data variable, the compost was frequently aerated for a different amount of duration each time throughout the 7 days.

The graph shows that the compost pile has an initial high moisture content of near 80%. However, as soon as the compost pile was aerated, the moisture content spreads, and the moisture level plunged to approximately 60%. The moisture level of the compost decreased from 60–45% approximately which is expected due to evaporation. A subtle increment of moisture content could be observed from the plot as well. This could be due to the water released from the aerobes as by-products during aerobic respiration or could also be the sensor’s sensitivity. Nonetheless, the moisture level of the compost shows promising results as matured compost has a moisture content of between 40–50%.

Observing the temperature plot, the temperature in the compost pile remained around the range of 30°C but showed a slight increment nearing the end of the 7th day. The temperature is expected to increase as microbes release carbon dioxide during aerobic respiration. However, the temperature in the compost pile did not show an obvious increment but fluctuated due to the frequent aeration as aeration would release heat stored in the compost pile. Despite the temperature not being in the optimum temperature range of 40°C to 65°C, the composting process still took place though matching the temperature in the optimum range would typically yield a faster composting process.

Compost soil transportation system

As shown in Fig. 8. The robot transporter was programmed to transfer the compost material to its destination by following a specified black tape route. The robot can turn left or right depending on the path.

Figure 9. Shows the robot transporter will come to a stop upon arrival at its destination. If the operator tells the robot to get another container of compost, it returns to its starting position.

In conclusion, the proposed intelligent kitchen waste composting system succeeded in achieving its objective to solve the kitchen waste issue efficiently with reduced human labor, cost, and time. All three subsystems were able to communicate with one another, making it a largely automated system. Based on the first subsystem results, the kitchen waste segregation system that used MobileNetV2 in its CNN architecture managed to classify the waste into compostable and non-compostable categories by obtaining a high F1 score of 0.8411. The composting system has a unique method of aeration without implementing an agitator to stir the compost but rotating the compost bin, leveraging gravitational force to flip the compost. The IoT system incorporated in the compost bin was created from scratch using MQTT IoT protocol and HTTP requests instead of relying on any open-source IoT platforms such as Blynk or ThingSpeak. Finally, the compost soil transportation system used PLA + for the robot's frame so it has high durability to transport bigger loads. The robot also successfully implemented IR sensor to produce an economical path tracking system. Nonetheless, the novelty of this project lies in the integration of all three subsystems together. Thus, the inauguration of the proposed intelligent kitchen waste composting system is expected to be able in catering the kitchen waste issues effectively with reduced human labour, cost, and time in the future.

This project was funded by UCSI University, Research Excellence & Innovation Grant REIG-FETBE-2022/026. The authors have no relevant financial or non-financial interests to disclose. The datasets generated during and/or analysed during the current study are not publicly available due to plagiarism possibility but are available from the corresponding author on reasonable request.

Abraham A, Ananthakrishnan GS, Verghese BB, Sivakumar S, Rakesh S (2020) SARGoT:Smart Autonomous Robotic Goods Transporter. Materialstoday: Proceedings 24(Part 3): 2030-2035. https://doi.org/10.1016/j.matpr.2020.03.633
Adabara I, Hiriji N, Emmanuel O, Alkasim SM, Zaina K, Mustafa MM (2019) Intelligent embedded agricultural robotic system. International Journal of Engineering and Information System 3(1): 14-24.
Ayilara MS, Olanrewaju OS, Bababola OO, Odeyemi O (2020) Waste management through composting: Challenges and potentials. Sustainability 12(11): 4456. https://doi.org/10.3390/su12114456
Bernama (2020) SWCorp: Food waste drops during MCO, rises again soon after. New Straits Times. https://www.nst.com.my/news/nation/2020/10/633738/swcorp-food-waste-drops-during-mco-rises-again-soon-after. Accessed 20 October 2020
Gubbi J, Buyya R, Marusic S, Palaniswami M (2013) Internet of Things (IoT): A vision, architectural elements, and future directions. Future Generation Computer Systems 29(7): 1645-1660. https://doi.org/10.1016/j.future.2013.01.010
Hanumante V, Roy S, Maity S (2013) Low cost obstacle avoidance robot. International Journal of Soft Computing and Engineering 3(4): 52-55. http://dx.doi.org/10.6084/M9.FIGSHARE.1327873
Kumar P, Ashok G (2021) Design and fabrication of smart seed sowing robot. Materialstoday: Proceedings 39(Part 1): 354-358. https://doi.org/10.1016/j.matpr.2020.07.432
Li R, Chen S, Li X, Lar JS, He Y, Zhu B (2009) Anaerobic codigestion of kitchen waste with cattle manure for biogas production. Energy Fuels 23(4): 2225-2228. https://doi.org/10.1021/ef8008772
MAEKO (2022) The food waste specialists. https://www.maeko.com.my/aboutus.php. Accessed 01 December 2022.
Nakanishi H, Shibata H (2017) Kitakyuhsu City, Japan. In: Roberts BH, Lindfield M, Steinberg F (eds) Partnerships for the sustainable development of cities on the APEC region, APEC Policy Support Unit, Asia-Pacific Economic Cooperation pp 183-201
Novianti PW, Tweel IVD, Long VL, Roes KCB, Eijkemans MJC (2015) An application of sequential meta-analysis to gene expression studies. Cancer Inform 14(Suppl 5): 1-10. https://doi.org/10.4137%2FCIN.S27718
Punetha D, Kumar N, Mehta V (2013) Development and Applications of Line Following Robot Based Health Care Management System. International Journal of Advanced Research in Computer Engineering & Technology 2(8): 2446-2450.
Secretariat, GMI (2022) U.S. Government Global Methane Initiative (GMI) Accomplishments: 2017 Annual Report. United States Environmental Protection Agency (EPA). https://www.epa.gov/gmi/us-government-gmi-accomplishments-2017-annual-report. Accessed 18 February 2022.
Svendsen JH, Diederichsen SZ, Hojberg S, Krieger DW, Graff C et al. (2021) Implantable loop recorder detection of atrial fibrillation to prevent stroke (The LOOP Study): a randomised controlled trial. The Lancet 398(10310): 1507-1516. https://doi.org/10.1016/S0140-6736(21)01698-6
Yao G, Lei T, Zhong J (2019) A review of convolutional-neural-network -based action recognition. Pattern Recognition Letters 118: 14-22. https://doi.org/10.1016/j.patrec.2018.05.018

GraphicalAbstract.png

Download PDF

Journal Publication

published 07 Dec, 2023

Read the published version in Waste and Biomass Valorization →

Reviewers agreed at journal
27 Jun, 2023
Reviewers invited by journal
25 Jun, 2023
Editor invited by journal
19 Jun, 2023
Editor assigned by journal
30 May, 2023
First submitted to journal
28 May, 2023

You are reading this latest preprint version

Intelligent Kitchen Waste Composting System via Deep Learning and IoT

Status:

Journal Publication

Version 1

Abstract

Figures

Highlights

Novelty

Introduction

System design overview

Result

Conclusion

Statements and Declarations

References

Supplementary Files

Status:

Journal Publication

Version 1