Page 86 - IJAMD-2-2

International Journal of AI for
Materials and Design Fruit image detection using AI

categorization without causing much harm to the fruits. 14-17 (i) Lighting conditions: Good lighting was vital in
In contrast, manual sorting of fruits by professional showing the visual traits of the fruits. An ESP32
personnel undergoes physical handling, thus potentially camera provided the best lighting and clarity.
damaging the fruits and affecting their value. 18-21 This (ii) Camera specifications: The camera’s resolution and
research focuses on creating an image processing system color accuracy had a big impact on image quality. We
for the classification of fruits through machine learning fine-tuned these factors to ensure the fruits’ features
to enhance precision and productivity in the agriculture were displayed accurately.
business as well as the food chain. Similar approaches
utilizing image processing and machine learning to detect 2.1.2. Pre-processing
mangoes, tomatoes, and oranges were also performed by Pre-processing prepared the captured images for feature
other researchers. Image processing methods and machine extraction by enhancing important properties and
learning algorithms have been widely used to classify reducing noise. The methods used include:
mangoes, tomatoes, and oranges, achieving classification (i) Resizing: The system resized all images to uniform
accuracies ranging from 80% to 100% across these fruit dimensions while preserving their aspect ratio. This
types. 22-37 Research on fruit classification using image step was essential to ensure consistency across datasets
processing and machine learning is constrained due to and to enhance computational efficiency.
its focus on single fruit type, small datasets, inconsistent (ii) Histogram equalization: This technique enhanced
image acquisition methods, and the lack of deep learning image contrast by spreading out pixel intensity values.
approaches. Future investigations should prioritize It standardized the appearance of images, making
standardized data collection and the application of deep key features more distinguishable during the later
learning algorithms. processing stage.
The proposed fruit classification system leverages (iii) Thresholding: Thresholding splits pixels into object
image processing to achieve maximum accuracy and and background areas based on a predefined value. It
minimal time expenditure. It is trained on a large database effectively isolated the fruit from the background and
comprising mangoes, tomatoes, and oranges. Such a reduced noise, thereby improving feature extraction.
classification system has significant potential to enhance 2.1.3. Feature extraction
the quality of both the agricultural sector and the food
industry by monitoring product quality, minimizing At this point, the system extracted various features
wastage, and adding value to the product. Furthermore, it from the pre-processed images. These features included
can be integrated into other automated systems and mobile color, shape, texture, and size, which were essential in
applications. differentiating fruits.
The novel integration of real-time image processing 2.1.4. Feature selection
with real-time mechanical fruit sorting, powered by After the features were extracted, the system evaluated the
artificial intelligence machine learning optimization significance of each feature. The features that varied widely
techniques using Python programming, represents the between fruit types were retained, while those with insignificant
novelty of this study.
variation were excluded. This step enhanced the sorting system
2. Materials and methods by focusing on the most distinctive features, which resulted in
better accuracy and reduced computational work.
2.1. Image processing system to process fruit
images 2.2. Support vector machine (SVM) model to classify
Four steps were included in developing the image fruit images
processing system to classify fruits (mango, oranges, and The SVM model helped classify fruits by their appearances.
tomatoes): Image acquisition, pre-processing, feature This part explained the key steps in building the SVM
extraction, and feature selection. Each step played a key model, which included standardizing the features, training
role in ensuring the accuracy and effectiveness of the the model, and applying different kernel functions.
classification system.
2.2.1. Standardizing the features
2.1.1. Image acquisition Standardizing the features was a key step before starting
This step was crucial to capture high-quality images of the the modeling process. It ensured that all the features
fruits, which we needed for later analysis. Crucial factors to selected for model creation contributed to the fruit-sorting
consider during image acquisition included: process. In this study, normalization was used to scale all

Volume 2 Issue 2 (2025) 80 doi: 10.36922/IJAMD025150011