Chapter III

METHODOLOGY

This chapter discusses and tackles the design and strategy employed by the researchers to

conduct and complete this study. The researchers have carefully planned the activities in order to

finish the study satisfactorily at a definite time. This chapter includes the research locale, research

design, data gathering procedure.

Research Design

This proposed project study of a tomatoes freshness level indicator is employing hybrid

neural networks to assess the freshness level of tomatoes through image classification. The

inspiration for using this hybrid approach systems from prior research findings highlighted in our

literature review, showcasing the success of neural networks in categorizing food.

Furthermore, they use a hybrid neural network because it combines the strengths of

different neural network architectures (Zhao R et al., 2022). For instance, they would be using

convolutional neural networks (CNNs) for feature extraction. Additionally, the researchers would

be using SVM, decision trees, and Random Forests for classification. The use of a hybrid neural

network approach allows us to create a more robust and accurate model for freshness level

classification (Zhao R et al., 2022). We introduce the following Hybrid Neural network algorithms

used in the classification of images; SVM+CNN, Decision Trees + CNN, and Random Forest +

CNN. They also list some of the advantages and disadvantages of using these algorithms.

“Convolutional Neural Network (CNN) is a class of deep feed-forward. artificial neural

networks which are commonly used in computer vision problems such as image classification”,

(Abien et al., 2017). A Support Vector Machine (SVM) + Convolutional Neural Network (CNN)
hybrid neural network is a combination of two powerful machine learning techniques used for

classification. tasks (Duggani Keerthana et al., 2022). CNN performs the task of extracting.

features from unprocessed data, such as images, whereas the SVM is utilized to classify the

extracted features. CNN first learns the features. From the input images, then the extracted features

are passed on to the SVM for classification. Szarvas et al., (2005) studied the automatically

optimized CNN. features and demonstrated that combining CNN with SVM could produce the

highest testing accuracy.

On the other hand, for cooling storage of the tomatoes, they will use tec1-12706. TEC1-

12706 12V 92W Thermoelectric Peltier Cooler Module is a semiconductor-based electronic

module that functions as a small heat pump. It transfers heat from the cooler side to the hotter side

by operating on the principle of the Peltier effect. They can be used in applications like cooling

computer CPUs, CCDs, portable refrigerators, medical supplies, and equipment

(quartz.component).

Data Collection: Gather a diverse dataset of tomato images at different freshness levels.

Preprocessing: Clean and standardize the dataset, ensuring uniformity in image resolution

and quality

Feature Selection: Identify relevant features within the images that contribute to freshness

assessment.

Model Selection: Choose a suitable machine learning model (e.g., convolutional neural

network - CNN) for image classification. Evaluate potential hybrid models that combine different

architectures for improved performance.

 Training: Divide the dataset into training, validation, and testing sets. Train the model using

the training set, adjusting parameters to optimize performance.

Evaluation: To fine-tune parameters, assess the model's performance on the validation set.

Validate the model on the testing set to simulate real-world scenarios.

Validation Methods: Implement cross-validation techniques to ensure model reliability.

Use metrics like accuracy and recall evaluating performance.

Interpretation: Analyze the model to understand its decision-making process. Identify

limitations in the model's performance.

Documentation: Document the entire process, including dataset details, model architecture,

and results. Provide insights into the strengths and weaknesses of the freshness level indicator.

Presentation: Communicate findings through visualizations, charts, and a comprehensive

report.

Share recommendations for potential improvements or future research directions.

Through this descriptive research design, you can develop a clear understanding of the

effectiveness and applicability of the freshness level indicator using machine learning for

tomatoes.

B. Research Project Development

 Developing a research project about a Tomato Freshness Level Indicator involves preliminary

investigation/concept, design, fabrication, testing, and evaluation.

B.1. Preliminary Investigation/Concept

Ripe tomatoes will spoil quickly, which is about 3-4 days of storage at room temperature, so

the shelf life of fresh tomatoes is relatively short without proper handling. The researchers found

out that the different types of tomatoes will spoil quickly. That is why we want to apply a cooling

system to the tomatoes, so that we can extend their shelf life. And by the used of machine learning

algorithms, we can easily designate the fresh tomatoes to not and identify their level of freshness.

The researcher will develop an analytical tool to indicate the freshness level of tomato with an

input of machine learning.

The farmers and distributors will have the better prices for their produce if they can assure

customers of its freshness with a clear freshness indicator, and also can better manage the logistics

of transporting and distributing tomatoes, minimizing losses due to spoilage. The researcher

purpose is to create a more accurate, automated, and efficient system for assessing the freshness

of tomatoes. Traditional methods may have limitations in terms of precision and speed, but with

the help of machine learning it can make a significant impact.

B.2. Design

The primary goal of the Tomato Freshness Level Indicator is to provide users with a reliable

and user-friendly tool to assess the freshness of tomatoes. The design should be intuitive,

accommodating various types of tomatoes, and able to adapt to different environmental conditions.

Parameters of Design
Sensors

Temperature sensor will measure the temperature inside the cooling system. Humidity

Sensor will measure the moisture inside the cooling system.

Machine Learning

Convolutional Neural Network (CNN) is responsible for extractions of image of a tomato.

Support Victor Machine (SVM) is for classifications of tomato based on the image extracted.

Microcontroller/Microprocessor

Processing power which selects a microcontroller with sufficient processing power to

handle data from multiple sensors and perform calculations for freshness assessment. Ensure

compatibility with chosen sensors and communication modules.

User Interface

Adding a display which implement a clear and easily readable display to present the

freshness level to users.

Calibration Mechanism

Adjustment controls will provide a simple and effective calibration mechanism to account

for variations in tomato types and user preferences. Calibration algorithm will develop an

algorithm that allows users to fine-tune freshness level thresholds.

Enclosure
 The materials use should be durable and food-safe material for the device enclosure. Ensure

the design allows for easy cleaning to maintain hygiene standards.

Safety Considerations

The researcher will ensure that materials used in the design comply with safety standards

for food-contact devices. Implement safety features to prevent electrical hazards.

Cost-Effective Manufacturing

The researcher will optimize material selection for cost-effectiveness without

compromising performance. Design for efficient and cost-effective manufacturing processes.

Scalability

The researcher will design the device with scalability in mind to accommodate mass production

if needed.

B.3. Fabrication

Tomato Freshness Level Indicator involves integrating various technologies and processes

to accurately assess the freshness of tomatoes. General overview of the fabrication processes:

1. Choosing appropriate sensors that can measure indicators of tomato freshness, such as

temperature sensor and humidity sensor.

2. Selecting a microcontroller or microprocessor to process data from the sensors. Arduino or

Raspberry Pi are common choices in this research project.

3. Designing a data acquisition system to collect information from the sensors.

 4. Developing algorithms to analyze the sensor data and determine the freshness level of the

tomatoes.

5. Implementing a display system to present the freshness level to the user. This could be a

simple LED indicator, an LCD display, or even a smartphone app if the system is more

complex.

6. Choosing an appropriate power supply system. This will be a direct power source if the

indicator is stationary.

7. Designing a protective enclosure for the electronics, ensuring it is durable, water-resistant

(if needed), and allows for easy access to components.

8. Rigorously test the system with various tomatoes at different freshness levels to ensure

accurate readings. Calibrate the system as needed to improve accuracy.

9. Designing a user interface that allows users to interact with the device, such as setting

preferences or calibrating the system.

10. Creating detailed documentation for users.

B.4. Testing

Testing is a crucial phase in the development of a Tomato Freshness Level Indicator to ensure

its accuracy, reliability, and performance.

Operation Procedure

The researcher will energize the machine and check all the components if it is all working.

Input the image of fresh and non-fresh tomato in the Convolutional Neural Network (CNN). Place

the tomatoes inside the cooling system and calibrate the temperature on its designated coolness.
Testing Procedure

The researcher will verify that the Support Victor Machine (SVM) are detecting color and

physical appearance accurately. Confirm that the microcontroller processes data correctly. Ensure

the display or output accurately represents the freshness level. Check the accuracy of freshness

level thresholds against known freshness standards.

Standards

The researcher will ensure compliance with relevant food safety standards, especially if the

device comes into direct contact with food. Comply with electrical safety standards to avoid

hazards. Consider any environmental impact and adhere to relevant standards.

Trials Test

The researcher will test the device with various types of tomatoes to ensure it is effective across

different varieties. Conduct tests in conditions that mimic real-world scenarios, such as varying

temperatures and humidity. Run the device continuously for an extended period to check for

stability and reliability.

B.5. Evaluation

Evaluating a Tomato Freshness Level Indicator involves assessing its performance against

predefined criteria to ensure accuracy, reliability, and practicality.

Evaluation Procedure:

The researcher will measure the accuracy of the sensors in detecting color and physical

appearance. Evaluate the effectiveness of the calibration process. Confirm that adjustments align

with expected changes in freshness. Assess the microcontroller's ability to process sensor data
accurately and in a timely manner. Verify that the freshness level displayed is clear and easily

understandable for users. Test the device with various types of tomatoes to ensure it performs

consistently across different varieties. Assess performance under real-world conditions, including

varying temperatures and humidity levels. Run the device continuously for an extended period to

check for stability and reliability over time.

C. Instrument and techniques used/instrumentation.

In developing a tomato freshness level indicator using machine learning, various

instruments, and techniques are involved. Here's an overview:

Instrument: Digital Camera or Imaging

Device Description: Capture high-resolution images of tomatoes at different freshness

levels.

Technique: Use a digital camera or imaging device to ensure consistent and accurate

representation of tomato characteristics.

Instrument: Image Preprocessing Tools

Description: Prepare and enhance the dataset for machine learning.

Technique: Utilize tools for resizing, cropping, normalizing, and augmenting images to

improve model robustness.

Instrument: Convolutional Neural Network (CNN) or Hybrid Models

Description: Machine learning models for image classification.

 Technique: Train a CNN or a hybrid model to recognize patterns and features associated

with different freshness levels in tomato images.

Instrument: Python Programming Language

Description: Widely used for machine learning and data manipulation.

Technique: Implement machine learning algorithms, libraries (such as TensorFlow or

PyTorch), and frameworks to develop and train the freshness level indicator.

Instrument: Python, Jupyter Notebooks, or Integrated Development Environment (IDE)

Description: Platforms for coding, testing, and documenting machine learning projects.

Technique: Use Python programs, jupyter notebooks or IDEs for coding the machine

learning model, running experiments, and documenting the code and results.

Instrument: Cross-Validation Techniques

Description: Assess model performance and generalization.

Technique: Employ techniques like k-fold cross-validation to ensure the model's reliability

and prevent overfitting.

Instrument: Evaluation Metrics

Description: Quantitative measures to assess model performance.

Technique: Use metrics such as accuracy, precision, recall, F1 score, and confusion matrix

to evaluate how well the freshness level indicator classifies tomatoes.

Instrument: Data Logging Tools

 Description: Tools for recording and documenting dataset details, model parameters, and

results.

Technique: Implement data logging tools to maintain a comprehensive record of the

development process, aiding in reproducibility and future improvements.

Instrument: Visualization Libraries

Description: Tools for presenting and interpreting results.

Technique: Utilize libraries like Matplotlib or Seaborn to create visualizations that help

understand model predictions and performance.

Instrument: Cloud Computing Services

Description: Resources for scalable and efficient model training.

Technique: Leverage cloud computing services like AWS, Google Cloud, or Azure to

handle large datasets and resource-intensive model training.

These instruments and techniques collectively contribute to the development and

evaluation of a freshness level indicator using machine learning for tomatoes.

D. Data Processing and Statistical Treatment of Data

Data Processing:

1. Data Collection: Gather a diverse dataset of tomato images with corresponding freshness

labels. Ensure a balanced representation of different freshness levels in the dataset.

2. Data Cleaning: Remove any irrelevant or duplicate images.

Address missing data and outliers to maintain dataset integrity.

 3. Data Splitting: Divide the dataset into training, validation, and testing sets. A common

split might be 70% for training, 15% for validation, and 15% for testing.

4. Image Preprocessing: Resize and standardize images to a consistent resolution.

Statistical Treatment of Data:

1. Descriptive Statistics: Calculate basic statistics (mean, median, standard deviation) for

numerical features related to image characteristics.

2. Label Distribution: Analyze the distribution of freshness labels in the dataset.

Ensure a balanced representation of each freshness level for model training.

3. Feature Importance: If applicable, assess the importance of different image features.

4. Error analysis: Investigate misclassifications and errors made by the model.

Identify patterns or trends in misclassifications for potential model improvement.

E. Schedule of Activities (Gantt Chart)

MONTHS→ 1ST 2nd 3rd 4th 5th 6th 7th 8th 9th 10th
SCHEDULE
PLANNING
RESEARCH
DESIGN AND
PROTOTYPING
TESTING
EVALUATION