Automated Trash Sorting System
Sumiyajid Ulziibayar
First Supervisor: Myagmarjav. B
Second Supervisor: Narangarav.T
November 24, 2025
The Global Waste Crisis
70%
Projected Growth
Global waste production increase by 2050 unless urgent action is taken
60%
Uncollected Waste
Garbage remains uncollected in low- and middle-income countries
7%
Mongolia's Recycling
Only 7% of Mongolia's 3 million tonnes of waste is recycled annually
Mongolia's Waste Challenge
Ulaanbaatar's Burden
The capital contributes nearly 60% of Mongolia's solid waste, with most ending up in open dumpsites like Nalaikh and Tsagaan Davaa landfills.
These sites pose serious environmental and health risks due to leachate contamination and spontaneous combustion.
Behavioral Barriers
Waste segregation is seen as a chore with little reward. Low participation rates stem from lack of incentives and sorting infrastructure.
The Nalaikh District proposed this project to modernize community waste handling through technology and behavioral design.
Project Objectives
01
Automated Classification
Design a system capable of identifying waste materials using computer vision with >90% accuracy
02
Behavioral Incentives
Establish a reward-based system to encourage consistent waste sorting through feedback and competition
03
Technical and economic feasibility
The final model should be easy to maintain with scalable design, and economically feasible.
System Scope
Three-Class Classification
Plastic Bottles, Metal Cans, and Others (paper, cardboard, glass, organic materials)
Precision Actuation
Stepper motor and servo mechanisms for reliable physical separation into designated bins
Data Integration
Camera-based vision system with high-precision weight sensor for enhanced accuracy
Reward System
Server-side framework for user tracking and group-based competition targeting school children
High-Level System Architecture
Raspberry Pi
CNN inference, image processing, main controller
ESP32
Motor control, sensor reading, actuation logic
Web Server
User tracking, reward system, leaderboards
The system uses modular components (Raspberry Pi, ESP32) to ensure scalability and easy replication in similar community settings.
Hardware Components
Raspberry Pi
Main computing unit for user interface, information updates, and upload
ESP32 Microcontroller
Handles motor control, sensor reading, and actuation commands
1080p USB Camera
Captures images for classification with integrated LED lighting
Load Cell
Measures item weight (5kg capacity) for enhanced classification accuracy and user progress tracking
NEMA 17 Stepper
Provides precise positioning for multi-bin sorting mechanism
Servo Motors
Controls trapdoors or latch mechanisms for bin access
Circuit Architecture
Electrical components and drivers
Raspberry pi
ESP-32
A4988
PCA9685
HX711
IR sensors
MPPT controller
Buck convertors
Voltage regulators
Power Management
  • 12V DC for stepper motors
  • 5V for Raspberry Pi and servos
  • 3.3V for ESP32 and sensors
  • Solar backup with 30Ah battery for outdoor deployment
Software Stack
1
Python (Raspberry Pi)
CNN model inference, image preprocessing, server communication, main control logic
2
C++ (ESP32)
Real-time motor control, sensor polling, serial communication handling, actuation sequences
3
Communication Protocols
Serial for Pi-ESP32, HTTP/HTTPS for server, JSON for structured data exchange
4
Web Application
MySQL database, user dashboard, classroom leaderboards, QR code activation system
Data Flow Process
1
QR code access
User interaction
2
Item Detection
IR sensor detects garbage
3
Data Capture
Camera captures image, load cell measures weight simultaneously
4
CNN Classification
Image sent to CNN model, returns classification result in JSON format
5
Actuation Decision
Raspberry Pi sends command to ESP32 based on classification result
6
Data update
The weight of the trash is then sent to the server for user progress tracking
System Requirements
Performance
  • Classification accuracy >90%
  • Inference time <1 second
  • Throughput: 10 items/minute
  • System uptime >95%
Environmental
  • Operating range: -30°C to +30°C
  • Dust/moisture resistant (IP54)
  • Insulated against freezing
  • Weatherproof enclosure
Connectivity
  • Wi-Fi (2.4 GHz) or cellular
  • Data transmission delay <3s
  • HTTPS encryption for security
AI Model & Training
Dataset Overview
Mongolian Waste Database
Custom dataset created specifically for local waste characteristics, addressing the gap in existing datasets.
Public contribution system via web interface allows continuous dataset growth and improvement.
  • 400+ images collected over one month
  • Three categories: Plastic, Metal, Others
  • Images preprocessed automatically upon upload
Initial Training Data
TrashNet dataset used for baseline training with six categories: cardboard, glass, metal, paper, plastic, and trash.
Thousands of labeled images provided foundation for transfer learning approach.
Data Preprocessing Pipeline for training
1
Image Resizing
All images standardized to 224×224 pixels for consistent model input
2
Normalization
Pixel values scaled between 0 and 1 to improve convergence
3
Data Augmentation
Random flipping, rotation, zoom, and brightness adjustments
4
Train/Val Split
80% training data, 20% validation, with separate test set
Augmentation techniques simulate real-world variations in lighting and object positions, making the model more robust to environmental interference.
CNN Architecture Explained
Convolutional Layer
Learnable filters scan input to detect spatial patterns like edges and textures
Activation Function
ReLU introduces non-linearity, enabling complex pattern recognition
Pooling Layer
Down-samples feature maps, reducing dimensions while preserving key features
Dense Layers
Fully connected neurons combine features for final classification
Softmax Output
Converts scores to probability distribution across waste categories
Convolutional Operations
The convolutional layer uses learnable filters (kernels) to scan the input image, generating output feature maps that detect spatial patterns.
Key Concepts
  • Local Connectivity: Each neuron connects only to a small region (receptive field)
  • Weight Sharing: Same filter weights used across all spatial positions
  • Translation Invariance: Detects features regardless of position in image
  • Multi-Channel Processing: Analyzes RGB channels simultaneously for color-dependent patterns
Benefits
Reduces learnable parameters compared to fully connected layers, improving computational efficiency and reducing overfitting risk.
Pooling Operations
Purpose
Pooling layers perform down-sampling to reduce spatial dimensions while maintaining essential features.
Max Pooling
Most commonly used function due to its ability to effectively map features within an image. Selects maximum value from each region.
Advantages
  • Reduces computational load
  • Provides translation invariance
  • Prevents overfitting
  • Maintains important features
Transfer Learning Approach
Base Models Tested
  • EfficientNetB0
  • EfficientNetV2 variants
  • ResNet50
  • MobileNetV2
All models pre-trained on ImageNet, enabling complex feature learning from extensive datasets.
Training Strategy
First half of base model layers frozen to preserve learned features. New classification head added for 6-class waste problem.
Adam optimizer with 1e-4 learning rate, categorical cross-entropy loss function, 15-30 epochs with early stopping.
Class weights computed to handle imbalanced dataset, ensuring fair representation of all waste categories.
Model Training Results
EfficientNetB0 achieved the highest classification accuracy of 91.97% on the test dataset, exceeding the 90% target. It offers the best trade-off between speed and accuracy for deployment.
Performance Metrics
91.97%
Test Accuracy
EfficientNetB0 model performance on held-out test set
0.91
F1-Score
Balanced measure of precision and recall across all classes
0.18
Validation Loss
Low loss indicates good model convergence and generalization
Precision and recall values exceeded 0.90 for main categories, confirming consistent performance. However, the model requires further training on Mongolian-specific waste data for reliable deployment.
Web Application Integration
QR Code Activation
When idle, Raspberry Pi displays QR code on screen. Scanning activates the system and prepares components for waste processing.
Direct camera activation available through web interface for seamless user experience.
Reward System Features
  • Interactive map showing all smart bin locations
  • Individual user contribution dashboard
  • Classroom/school leaderboard rankings
  • Real-time statistics and progress tracking
Behavioral Design Strategy
Competition
Group-based leaderboards motivate consistent participation
Gamification
Points and achievements make sorting engaging and rewarding
Social Recognition
Visible rankings provide peer motivation and accountability
Habit Formation
Targeting children creates long-term behavioral change
Progress Tracking
Real-time feedback reinforces positive sorting behavior
Database Structure
1
Users & Classrooms
Tracks individual contributions and classroom affiliations. Links users to their total recycling weight and classroom rankings.
2
Devices & Bins
Maintains device locations, serial numbers, and bin configurations. Monitors device health and last connection times.
3
Reports
Records every disposal event with classification, weight, image URL, and user ID. Enables analytics and leaderboard generation.
4
Misclassifications
Stores human corrections of AI errors. Creates feedback loop for continuous model improvement.
Mechanical Design Requirements
Structural Integrity
Robust construction to withstand outdoor conditions, resisting corrosion, impact, and fatigue for long-term operation.
Precise Actuation
Accurate sorting mechanisms with controlled movement for reliable separation of different waste categories.
User Safety & Accessibility
Features to prevent injury during interaction, including clear interfaces and easily accessible emergency stop mechanisms.
Weather Resistance
Enclosures and components rated for extreme temperatures, precipitation, and UV exposure for reliable outdoor deployment.
Maintenance Accessibility
Modular design allowing easy access to internal components for quick repairs, cleaning, and preventative maintenance.
Size Constraints
Optimized dimensions for efficient space utilization in various urban and rural deployment scenarios.
Mechanical Configuration
The input bin and integrated weighing system underwent extensive design and testing through multiple iterations.
  1. 1 moving stepper design
  1. 2 servo design
  1. 1 stepper with multi servo design
  1. 1 servo 1 stepper design
  1. 1 servo with 2 stepper design
This iterative approach allowed us to refine the mechanical configuration for optimal performance, durability, and user interaction, resulting in the current design of the prototype.
Figure shows a 3D design drawn on cad.onshape.com that was not completed
Other smaller parts were 3D printed uniquely custom to the project.
Key Achievements
Classification Success
CNN model achieved 91.97% accuracy, exceeding the 90% target requirement
Dataset Creation
Established first public Mongolian waste image database with 400+ labeled images
Modular Architecture
Developed scalable system supporting multiple deployment scenarios and configurations
Web Platform
Created functional reward system with real-time tracking and classroom leaderboards
ESP-32 code functions
Main Functional Flow:
01
Lift up
02
Measure weight
03
Lower lift
04
Move box
05
Open latch
06
Return to positions
System Features & Control Interface
Weighing Functionality
  • Automated lift sensor positioning for precise measurements
  • Real-time weight sampling with HX711 and data averaging
  • Tare functionality for empty box weight calibration
Motor Control
  • Non-blocking stepper motor movements for simultaneous operations
  • Horizontal box positioning and vertical lift control
  • Precise positioning with step-based movement tracking
Demo Mode
  • Automated sequence: move box → unlock latch → wait → lock latch → return
  • Interactive serial command interface for manual control
  • Status reporting for system state monitoring
Current Limitations with future improvements
Partial Integration
System modules functional individually but not yet assembled into unified physical device due to late hardware arrival.
Dataset Size
Mongolian waste dataset still limited (400 images). Model requires retraining with more local data for reliable deployment.
Field Testing
No long-term operational validation yet. System performance under continuous real-world use remains to be evaluated.
Mechanical Design
Enclosure manufacturing and full mechanical assembly pending. Weather protection for outdoor deployment not yet implemented.
Electrical configuration
The final product will be using a custom printed PCB with IR sensors and Limit switches, and possibly other sensors such as NIR.
Connectivity
The model will be recieving A7670E SIM module for LTE connection incase of no WiFi availibility.
Future Work plan until June
Dataset Expansion
Continue collecting Mongolian waste images through public web interface. Target 2000+ images for robust model retraining.
Physical Assembly
Complete hardware integration into cohesive prototype. Manufacture weatherproof enclosure for outdoor deployment.
GMIT Pilot
Deploy system at GMIT to observe user behavior, system reliability, and long-term data consistency.
Optimization
Enhance communication stability, power efficiency, and cloud scalability for broader deployment across schools.
Thank You
This project demonstrates how engineering, AI, and human-centered design can converge to tackle Mongolia's urban waste challenges.
"The Automated Trash Sorting System stands not only as a technical prototype but also as a stepping stone toward building a culture of responsible waste handling in Mongolia."
Questions?