AI-Based Waste Classification and Smart Waste Management Systems: Techniques, Challenges, and Future Directions

AI-Based Waste Classication and Smart Waste Management Systems: Techniques, Challenges, and Future Directions

Mohammed Zain I

HKBK College of Engineering, Bengaluru, India

Mohammed Hashir

HKBK College of Engineering, Bengaluru, India

Mohammad Zuhaib Wani

HKBK College of Engineering Bengaluru, India

Md Yusuf Ali

HKBK College of Engineering Bengaluru, India

Kavitha D D

HKBK College of Engineering Bengaluru, India

Abstract – Rapid urban growth and the corresponding rise in municipal solid waste have intensied the need for efcient and scalable waste management solutions. Conventional segregation practices, which largely rely on manual intervention, are often time-consuming, inconsistent, and unsuitable for handling large volumes of waste. In recent years, advancements in Articial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) have facilitated the development of intelligent systems capable of automating waste classication and monitoring processes.

This survey provides a detailed review of 13 research studies published between 2021 and 2025, with a focus on AI-driven approaches for waste management. It examines a range of techniques, including Convolutional Neural Networks (CNNs), object detection frameworks, and hybrid learning models, along with the datasets, evaluation metrics, and system designs employed in these works. A comparative assessment is carried out to analyze the performance, advantages, and limitations of the existing methodologies.

The review further identies critical research gaps, such as the absence of standardized benchmark datasets, limited real- time implementation, lack of multimodal sensor integration, and insufcient emphasis on model interpretability. Based on these ndings, the paper outlines potential future directions, including the adoption of edge-based AI systems, integration with IoT-enabled smart infrastructure, and the use of advanced detection models to support scalable and efcient waste management in smart city environments.

KeyWordsWaste Segregation, Deep Learning, Convolutional Neural Networks (CNN), Object Detection, Smart Cities, Inter- net of Things (IoT), Articial Intelligence

1. INTRODUCTION

   1. A. Background: The continuous growth of urban pop- ulations, industrialization, and changing consumption pat- terns has resulted in a substantial rise in municipal solid waste across the globe. Inefcient handling of this waste contributes to serious environmental concerns, including soil degradation,. water contamination, and increased greenhouse gas emissions. Effective segregation at the source plays a cru- cial role in addressing these issues, as it facilitates recycling, reduces reliance on landlls, and supports sustainable waste management practices. However, conventional approaches

      largely depend on manual sorting, which is not only time- intensive but also susceptible to inconsistencies and human error, especially in large-scale scenarios.

      Fig. 1: AI-based waste classication pipeline from image capture to smart bin segregation

   2. B. Problem Statement: Manual waste segregation systems face multiple operational challenges that limit their effectiveness. These include inaccurate classication, mix- ing of recyclable and non-recyclable materials, and higher labor costs. In densely populated urban areas, the diverse nature of waste further complicates the segregation process, ultimately affecting recycling outcomes and environmental safety. Moreover, traditional systems lack the capability to monitor waste conditions in real time or adapt dynamically, making them inefcient for modern waste management re- quirements. These challenges underline the necessity for intelligent and automated solutions that can operate reliably at scale. As shown in Figure ??,1 the system processes waste images in real time.

   3. C. Role of Articial Intelligence: Advancements in Articial Intelligence (AI), along with Machine Learning (ML) and Deep Learning (DL), have signicantly enhanced the capabilities of image-based analysis and object recogni- tion systems. Among these, Convolutional Neural Networks

      (CNNs) have proven particularly effective for classication tasks due to their ability to automatically extract and learn hi- erarchical features from visual data. Such models are widely applied in waste segregation systems to identify categories like plastic, metal, paper, and organic materials.

      In addition to conventional CNN architectures, more ad- vanced models such as ResNet and EfcientNet have demon- strated improved accuracy and efciency, making them suit- able for complex classication tasks. Furthermore, object detection frameworks like YOLO enable real-time identica- tion of multiple waste items within a single frame, which is highly benecial for applications such as automated sorting and smart waste bins. These developments have transformed traditional classication methods into more intelligent and responsive systems capable of real-time operation.

   4. E. Challenges in Current Research: Despite notable progress, several limitations remain in existing intelligent waste management solutions. A key concern is the absence of standardized datasets, as many studies rely on limited or custom data, making it difcult to benchmark and compare different approaches. Additionally, a signicant portion of research is conducted under controlled environments, which does not fully capture real-world conditions such as lighting variations, occlusion, and complex backgrounds.

      Another challenge lies in the computational demands of deep learning models, which can hinder deployment on resource-constrained edge devices. Furthermore, many sys- tems lack interpretability, making it difcult to understand model decisions and establish user condence. The limited use of multimodal approaches, which combine visual data with sensor inputs, also restricts the robustness and adapt- ability of current systems.

   5. F. Motivation: The increasing need for sustainable and scalable waste management solutions, combined with the shortcomings of existing methods, motivates this study. There is a clear demand for a structured analysis that not only reviews current techniques but also evaluates their practical applicability and limitations. By identifying research gaps and highlighting emerging trends, this survey aims to provide meaningful insights that can guide future developments in intelligent waste management systems.

2. Contribution of This Survey

   This survey presents a detailed and structured analysis of 25 research works, including 13 IEEE publications and foundational deep learning studies. The major contributions of this paper are summarized as follows:

   1. A comprehensive review of AI-based waste man- agement systems published between 2021 and 2025 [1][13].

   2. Analysis of various algorithms including CNN, ma- chine learning models, and real-time object detection techniques.

   3. Evaluation of datasets, input-output frameworks, and performance metrics used in existing research.

   4. Identication of key research gaps such as lack of standardization, scalability issues, and absence of ex- plainable AI.

   5. Discussion of technical challenges associated with real-world deployment and edge computing.

   6. Proposal of future research directions including mul- timodal sensing, edge AI optimization, and smart city integration.

   The ndings of this survey provide a strong foundation for the development of next-generation intelligent waste management systems that are scalable, efcient, and envi- ronmentally sustainable. This survey presents a comprehen- sive overview of recent advancements in intelligent waste management systems driven by Articial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL). With the rapid increase in urban population and waste gener- ation, traditional waste segregation methods have become inefcient and unsustainable. This study analyzes multiple research works published between 2021 and 2025, focusing on automated waste classication and monitoring techniques.

   Fig. 2: Distribution of AI Techniques Used in Reviewed Papers

   The paper examines various approaches, including Convo- lutional Neural Networks (CNNs), general machine learning models, and computer vision-based systems. CNN-based methods are widely adopted due to their effectiveness in image classication tasks, particularly in identifying different types of waste such as plastic, paper, metal, and organic materials. Additionally, advanced architectures and object detection models have enabled more accurate and real- time waste identication, improving system efciency. Some studies also integrate Internet of Things (IoT) technologies with AI models to enable smart monitoring, real-time data collection, and optimized waste management processes.

   A comparative analysis of selected research papers high- lights the strengths and limitations of existing approaches. While many models achieve high classication accuracy, sev- eral challenges remain. These include the lack of standard- ized datasets, limited validation in real-world environments, and high computational requirements that restrict deployment on edge devices. Furthermore, most systems rely solely

   on image data and do not incorporate multimodal inputs such as sensor data, which could enhance performance and reliability.

3. MATH

   Before you begin to format your paper, rst write and save the content as a separate text le. Keep your text and graphic les separate until after the text has been formatted and styled. Do not use hard tabs, and limit use of hard returns to only one return at the end of a paragraph. Do not add any kind of pagination anywhere in the paper. Do not number text heads-the template will do that for you.

   COMPARATIVE ANALYSIS

   Despite signicant advancements, several research gaps re- main in the domain of intelligent waste management systems. First, there is no standardized dataset available for waste classication. Most studies rely on custom datasets, leading

   to inconsistencies in performance evaluation.

   Second, limited real-time deployment is observed. While some studies propose real-time systems, very few validate them in real-world environments.

   Third, the absence of multimodal sensing is evident. Current systems rely primarily on image data, ignoring other important factors such as weight, moisture, and chemical composition.

   Fourth, explainable AI techniques are not widely adopted. Most deep learning models operate as black boxes, reducing transparency and trust.

   Fifth, scalability remains a major challenge. Existing sys- tems are tested in controlled environments and may not perform effectively in large-scale smart city deployments.

   Finally, energy efciency and computational cost are rarely considered, despite being critical for real-time and edge- based systems.

   The comparative evaluation of existing research in intelli- gent waste management systems reveals several important observations regarding the effectiveness, limitations, and practical applicability of different approaches. A detailed analysis of the reviewed studies indicates that deep learning- based methods, particularly Convolutional Neural Networks (CNNs), dominate the eld due to their superior performance in image-based classication tasks. These models are capable of automatically extracting complex and hierarchical features from input data, enabling accurate identication of various waste categories such as plastic, metal, paper, and organic materials.

   In contrast, traditional machine learning techniques, while computationally less intensive, generally exhibit lower ac- curacy and limited adaptability when compared to deep learning models. Although these approaches are easier to implement and require fewer resources, they often fail to capture complex visual patterns, which reduces their effec- tiveness in real-world waste classication scenarios.

   Recent advancements also highlight the integration of Arti- cial Intelligence with Internet of Things (IoT) technologies, leading to the development of smart waste management sys- tems. These systems leverage sensor data along with image- based inputs to enable real-time monitoring and decision- making. However, despite their potential, such systems are still in the early stages of deployment and face challenges related to scalability and infrastructure requirements.

   One of the major limitations observed across the reviewed literature is the absence of standardized datasets. Most stud- ies rely on custom-built datasets, which vary in size, quality, and diversity. This lack of uniformity makes it difcult to perform fair comparisons between different models and limits the generalization capability of trained systems.

   Another critical issue is the limited validation of proposed systems in real-world environments. Many studies demon- strate high accuracy under controlled laboratory conditions but do not account for practical challenges such as varying lighting conditions, occlusion, background noise, and mixed waste scenarios. As a result, the real-world performance of these models remains uncertain.

   Furthermore, the majority of existing systems rely solely on image data, neglecting the potential benets of multi- modal sensing. The integration of additional parameters such as weight, moisture content, and gas emissions could signi- cantly enhance classication accuracy and system reliability. The lack of such multimodal approaches represents a key research gap in the current literature.

   Explainability is another area that requires attention. Most deep learning models operate as black-box systems, mak- ing it difcult to interpret their decision-making processes. This lack of transparency can reduce user trust and hinder adoption in critical applications.

   Finally, computational complexity and energy efciency are important challenges that are often overlooked. Many deep learning models require high processing power, which limits their deployment on low-resource edge devices. For practical implementation in smart city environments, there is a need for lightweight and energy-efcient models that can operate in real time.

   Overall, while signicant progress has been made in the development of AI-based waste management systems, several challenges must be addressed to achieve scalable, efcient, and real-world deployable solutions

   A. Technical Challanges

4. Technical Challenges

The design and deployment of intelligent waste man- agement systems present a range of technical challenges that must be addressed to ensure reliable and scalable operation. One of the primary challenges is environmen- tal variability. In real-world scenarios, waste classica- tion systems are exposed to diverse and unpredictable

TABLE I: Summary of Reviewed Research Papers

TABLE II: Comparative Analysis of Waste Management Techniques

conditions, including uctuations in lighting, complex backgrounds, and partial occlusion of objects. These factors can signicantly degrade the performance of vision-based models, particularly when they are trained on limited or controlled datasets.

Another critical issue is class imbalance within available datasets. In many cases, certain categories of waste, such as hazardous or electronic waste, are underrep- resented compared to more common classes like plastic or paper. This imbalance can lead to biased model pre- dictions, where the system becomes more accurate for dominant classes while performing poorly on minority categories. Addressing this issue requires careful dataset curation, augmentation techniques, and the use of robust training strategies.

Real-time processing is another major concern, espe- cially for applications involving smart bins and auto- mated sorting systems. Deep learning models, while highly accurate, often demand substantial computational resources, making it challenging to deploy them on low- cost hardware platforms. This limitation becomes more pronounced in edge computing environments, where devices are constrained by limited memory, processing power, and energy availability.

Edge deployment further introduces the challenge of optimizing models without compromising performance. Techniques such as model compression, pruning, and quantization are often required to reduce computational overhead while maintaining acceptable accuracy levels. Achieving this balance between efciency and perfor- mance remains a key research problem.

In addition, the adaptability of these systems across different geographical regions poses a signicant chal- lenge. Waste composition varies widely depending on

local consumption patterns, cultural practices, and eco- nomic conditions. As a result, models trained on one dataset may not generalize effectively to another region. This necessitates continuous model updating, retrain- ing, and the development of more generalized learning frameworks.

Overall, addressing these technical challenges is es- sential for the successful implementation of intelligent waste management systems in real-world environments. Future research must focus on improving model ro- bustness, enhancing dataset diversity, and developing efcient algorithms that can operate under practical constraints.

• V. Challenges and Research Gaps in Mixed

Waste Processing

Processing and segregating waste in real-world envi- ronments becomes signicantly more complex when dealing with clustered or mixed waste. Unlike controlled scenarios where individual waste items are clearly visible, practical waste streams often contain overlap- ping, partially hidden, and contaminated materials. This introduces several technical challenges that limit the effectiveness of current AI-based systems.

One of the primary challenges is object overlap and occlusion. In mixed waste scenarios, multiple objects may be stacked or partially hidden, making it difcult for vision-based models to accurately detect and clas- sify each item. Traditional classication models, which assume a single object per image, are not well-suited for such conditions. Even advanced object detection models may struggle when objects are tightly clustered or visually similar.

Fig. 3: System Architecture

Another signicant challenge is the presence of hetero- geneous materials within a single frame. Waste items often differ in shape, texture, and color, and may also be deformed or contaminated. This variability increases the complexity of feature extraction and reduces clas- sication accuracy. Additionally, background noise and clutter further complicate the detection process, espe- cially in unstructured environments.

Real-time processing of clustered waste is also com- putationally demanding. Detecting multiple objects si- multaneously requires more complex models and higher processing power, which can be difcult to achieve on edge devices. This creates a trade-off between accuracy and efciency, particularly in smart bin or conveyor- based systems.

From a research perspective, several gaps remain un- addresed. Most existing studies focus on single-object classication using clean and well-separated datasets, which do not reect real-world conditions. There is a lack of large-scale datasets specically designed for mixed waste scenarios, including images with overlap- ping and partially occluded objects.

Furthermore, limited research has been conducted on

instance-level segmentation techniques, such as Mask R-CNN, which can distinguish individual objects within a cluster. The integration of multimodal sensing, com- bining visual data with depth sensors or weight mea- surements, is also largely unexplored in this context. Another important gap is the absence of robust eval- uation metrics for clustered waste scenarios. Current metrics primarily focus on classication accuracy and do not adequately measure performance in detecting multiple objects within a single frame.

Addressing these challenges requires the development of advanced detection and segmentation models, im- proved dataset diversity, and efcient algorithms ca- pable of real-time processing. Future research should focus on designing systems that can accurately identify and separate mixed waste in complex environments, enabling practical deployment in large-scale smart waste management systems.

VI. Future Research Directions

Future research in intelligent waste management sys- tems should prioritize the development of standardized and publicly available datasets. The absence of uniform datasets currently limits the ability to fairly compare different models and evaluate their generalization ca- pabilities. Creating large-scale, diverse datasets that capture real-world conditions will signicantly improve benchmarking and model reliability.

Another important direction is the integration of multi- modal sensing techniques. Combining image-based data with additional inputs such as weight, moisture levels, and chemical composition can enhance classication accuracy and provide a more comprehensive under- standing of waste characteristics. Such approaches can lead to more robust and reliable systems, especially in complex environments involving mixed or clustered waste.

The optimization of deep learning models for edge deployment is also a critical research area. Techniques such as model compression, pruning, and quantization should be further explored to reduce computational overhead while maintaining performance. This will enable efcient deployment on low-power embedded devices, which are essential for real-time applications in smart bins and automated sorting systems.

Furthermore, there is a need to validate proposed systems in large-scale, real-world environments. Many existing studies are limited to laboratory settings, which do not fully capture the challenges of practical deploy- ment. Future systems must be tested under dynamic conditions to ensure scalability and reliability in smart city infrastructures.

Explainable Articial Intelligence (XAI) should also be incorporated into waste management systems to improve transparency and interpretability. Providing in- sights into model decisions can enhance user trust and facilitate adoption in critical applications.

In addition, future systems should extend beyond classi- cation tasks to include waste lifecycle analysis, logis- tics optimization, and environmental impact assessment. This holistic approach can support more sustainable and efcient waste management strategies.

Finally, the adoption of advanced algorithms such as real-time object detection models (e.g., YOLO), transformer-based architectures, and reinforcement learning techniques can further improve system adapt- ability and performance in complex scenarios.

Further research can also explore the use of federated learning approaches to enable collaborative model training without sharing sensitive data across multiple locations. This can improve model generalization while maintaining data privacy. In addition, the incorporation of adaptive learning mechanisms can allow systems to continuously update and improve based on new waste patterns and environmental conditions. The integration of robotics with AI-based classication systems presents another promising direction, enabling fully automated waste sorting in industrial settings. Finally, the development of standardized evaluation frameworks and performance benchmarks will be essential to ensure consistency, reliability, and fair comparison across future intelligent waste management solutions. Moreover, future systems can benet from the integration of digital twin technology to simulate waste management processes and optimize system performance before deployment. The use of cloud-edge hybrid architectures can further improve scalability and responsiveness, enabling efcient data processing and decision-making across distributed smart waste management networks.

Table III summarizes the key limitations identied in existing studies, highlighting the need for more scalable, real-time, and interpretable waste management solutions.

VII. Conclusion

This survey presents a detailed review of recent devel- opments in intelligent waste management systems, with a focus on research published between 2021 and 2025. The analysis reects a clear transition from conven- tional machine learning techniques to more advanced deep learning models and integrated real-time solutions. These advancements have signicantly improved the accuracy and efciency of waste classication systems, particularly in image-based applications.

Despite these improvements, several challenges con- tinue to limit practical deployment. Issues such as scalability, high computational requirements, and lack of real-world validation remain critical barriers. In ad- dition, the absence of standardized datasets makes it difcult to benchmark and compare different approaches effectively. These limitations highlight the gap between theoretical research and real-world implementation.

The ndings of this survey suggest that future work should emphasize the development of robust, scalable, and energy-efcient systems capable of operating in dy- namic environments. Integrating Articial Intelligence with Internet of Things (IoT) technologies and smart infrastructure can further enhance system capabilities and enable real-time decision-making.

Overall, this study provides a consolidated understand- ing of current methodologies, identies key research challenges, and outlines potential directions for future work. It is expected that the insights presented will support ongoing efforts toward the development of practical and sustainable intelligent waste management solutions.

Acknowledgment

The authors would like to express their sincere gratitude to the faculty of the Department of Computer Science and Engineering, HKBK College of Engineering, Ben- galuru, for their guidance and support throughout this work.

References are important to the reader; therefore, each citation must be complete and correct. If at all possible, references should be commonly available publications.

References

1. M. R. S. Sandhya Devi, V. R. Vijay Kumar, and P. Sivakumar, “A Review of Image Classication and Object Detection on Ma- chine Learning and Deep Learning Techniques,” in Proc. 5th Int. Conf. Electronics, Communication and Aerospace Technology (ICECA), Coimbatore, India, 2021, pp. 18.

2. K. B., S. R., and P. M., “Enhanced CNN Architecture for Accurate Waste Classication in Smart Cities,” in Proc. Int. Conf. IoT Based Control Networks and Intelligent Systems (ICICNIS), Bengaluru, India, 2024, pp. 538543.

3. R. L. Kumar, R. Ramya, M. J. Balaji, V. Hari, and M. Malarvizhi, “Garbage Collection and Segregation using Com- puter Vision,” in Proc. Int. Conf. Inventive Computation Tech- nologies (ICICT), Lalitpur, Nepal, 2024, pp. 10231028.

4. P. Prasanth and R. Raut, “Anayzing and Categorizing Waste Using Convolutional Neural Networks and TensorFlow,” in Proc. 2nd World Conf. Communication Computing (WCONF), Raipur, India, 2024, pp. 16.

5. M. Kumar, S. Kumar, A. K. Luhach, and A. Tiwari, “Advance- ments in Waste Segregation Through Machine Learning and Integrating AI for Sustainable Waste Management,” in Proc. 13th Int. Conf. System Modeling Advancement in Research Trends (SMART), Moradabad, India, 2024, pp. 328334.

6. A. V., B. Nithya, G. M., B. T. M., A. Hiremath, and B. Jayaprakash, “Ingenious Waste Management: Implementing Machine Learning in Waste Segregation,” in Proc. Int. Conf. Emerging Smart Computing and Informatics (ESCI), Pune, India, 2024, pp. 15.

7. A. S. M., S. Ashwath, A. V. Gadagkar, and S. S., “Intelligent Waste Sorting System Using Deep Learning,” in Proc. IEEE Int. Conf. Blockchain and Distributed Systems Security (ICBDS), Pune, India, 2024, pp. 16.

8. M. R. O., S. B. D., V. Vivek, K. Prathik, N. Nishmitha, and

   S. Bhat, “Design and Testing of an AI-based Campus Waste Management System,” in Proc. Int. Conf. Intelligent Systems and Pioneering Innovations in Robotics and Electric Mobility (INSPIRE), Mangalore, India, 2025, pp. 477481.

9. P. D. M., P. A., S. N. V., S. B., G. V., and S. T., “Smart Waste

   Management System Using Machine Learning,” in Proc. 6th Int. Conf. Inventive Research in Computing Applications (ICIRCA), Coimbatore, India, 2025, pp. 18801885.

   TABLE III: Mapping of Research Gaps in Existing Studies

   Gap	Description	Papers Affected
   No Standard Dataset	Use of different custom datasets leading to inconsis- tent evaluation	[2][13]
   No Real-Time Systems	Majority of models operate ofine without real-time implementation	[2][7], [11], [12]
   No IoT Integration	Lack of sensor-based data integration in most sys- tems	[2][7], [9], [11]
   No Explainable AI	Models behave as black boxes with no interpretabil- ity	All
   High Computational Cost	Deep learning models require high processing power	[7], [10], [13]
   Limited Scalability	Systems tested only in small or controlled environ- ments	Majority of studies

10. S. S., W. A. Breen, M. Narmadha, K. S., and T. Bernatin, “Intelligent Waste Monitoring and Classication System with Automated Disposal Suggestions,” in Proc. 5th Int. Conf. Soft Computing for Security Applications (ICSCSA), Salem, India, 2025, pp. 21142119.

11. L. S. Ramesh, J. D. R., and K. S., “Waste Management System Using Deep Learning,” in Proc. Int. Conf. Recent Innovation in Science Engineering and Technology (ICRISET), Chennai, India, 2025, pp. 18.

12. B. Vishnu, D. Tanuj, and G. Saranya, “Smart Waste Man- agement: A CNN-based Approach for Automated Plastic and Material Classication,” in Proc. 5th Int. Conf. Ubiquitous Computing and Intelligent Information Systems (ICUIS), Erode, India, 2025, pp. 778783.

13. A. Gore, A. Bishoi, and R. S. Kute, “AI-Powered Waste Management: Real-Time Garbage Detection and Categoriza- tion,” in Proc. 12th Int. Conf. Emerging Trends in Engineering Technology – Signal and Information Processing (ICETET-SIP), Nagpur, India, 2025, pp. 15.

14. H. Yang and J. Thung, “Classication of Trash for Recyclability Status,” CS229 Project Report, Stanford University, 2016.

15. M. Yang and G. Thung, “TrashNet: A Dataset for Garbage Classication,” Stanford AI Project, 2016.

16. K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770778.

17. M. Tan and Q. Le, “EfcientNet: Rethinking Model Scaling for Convolutional Neural Networks,” in Proc. Int. Conf. Machine Learning (ICML), 2019.

18. J. Redmon et al., “You Only Look Once: Unied, Real-Time Object Detection,” in Proc. IEEE CVPR, 2016.

19. A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, “YOLOv4: Optimal Speed and Accuracy of Object Detection,” arXiv preprint arXiv:2004.10934, 2020.

20. G. Jocher et al., “YOLOv5 by Ultralytics,” GitHub Repository, 2020.

21. A. Howard et al., “MobileNets: Efcient Convolutional Neu- ral Networks for Mobile Vision Applications,” arXiv preprint arXiv:1704.04861, 2017.

22. S. Albawi, T. A. Mohammed, and S. Al-Zawi, “Understanding of a Convolutional Neural Network,” in Proc. Int. Conf. Engi- neering and Technology (ICET), 2017.