Computer Vision Project - Fruits & Vegetables Classifier

Our goal for this project was to create a PyTorch based computer vision model, dedicated to identifying & classifying a large variety of fruits & vegetables.

To achieve this goal, our project will include all the necessary steps including creating and training a CNN-based deep learning model, working with a large dataset of labeled images, handling input images for classification and testing.

** the notebooks in this project are fully documented **

Example of the model prediction based on an mp4 video input containing images from the testing dataset:

Libraries

• os
• torch
• torchvision
• tarfile
• matplotlib
• cv2 (openCV)

Tools

• Dataset: “Fruits 360”
  https://www.kaggle.com/moltean/fruits
• Deep Learning Framework: Pytorch
• Anaconda
• Python 3.9.7
• Jupyter notebooks/Visual Studio Code

Further Improvements

• Condition & Quality - It’s possible to improve our model by creating a dataset containing labeled images of fruits/vegetables with different conditions, defects or damage, and teaching our model to identify and grade the quality of fruits/vegetables in addition to classification abilities.
• Video Classification - although we successfully managed to implement video classification feature in out project, it is far from perfect. There are a lot of improvements to be made, such as live boundering boxes, predicting multiple images in a frame, and faster overall performance.
• Diversify the dataset - as mentioned, our experiments led to the conclusion that the dataset is not diverse enough, hence the need to diversify the data set for a more flexible and capable model predictions.

Bibliography

“Building a Deep Learning model with Pytorch to classify fruits and vegetables”
https://medium.com/swlh/building-a-deep-learning-model-with-pytorch-to-classify-fruits-and-vegetables-30e1a8ffbe8c

“Build an Image Classification Model using Convolutional Neural Networks in PyTorch”
https://www.analyticsvidhya.com/blog/2019/10/building-image-classification-models-cnn-pytorch/

Fruits 360 dataset by “kaggle”
https://www.kaggle.com/moltean/fruits

“Implementing Real-time Object Detection System using PyTorch and OpenCV”
https://towardsdatascience.com/implementing-real-time-object-detection-system-using-pytorch-and-opencv-70bac41148f7 https://towardsdatascience.com/object-detection-and-tracking-in-pytorch-b3cf1a696a98