Training-Machine-learning-models-for-weed-classification

This Image classification task (Weed classification) is very important in agriculture as it helps farmers identify and manage unwanted plant species that can compete with crops for resources, leading to reduced yields and economic losses. Traditional methods of weed identification and management takes longer time and requires more effort. Advancement in computer vision technology has made this process easy there by reducing labour costs, raising productivity, and ensuring the sustainability of agricultural practices. The main aim of this project is to train different machine learning-based classifiers using image processing techniques to accurately classify two weed classes: charlock and cleavers.

Background information

The dataset was adjusted from the plant seedling classification competition on Kaggle. Subset of charlocks and cleavers images was downloaded and stored in a folder named Data Image processing weed classification having three sub-folders which contains the test, train and validation folder, each subfolder contains two different folders named charlocks and cleavers which contains the images of the charlocks and cleavers weed respectively.

• Total number of images in the dataset = 786 images of both charlock and cleavers
• Total number of images in the train folder = 480 images of both charlock and cleavers
• Total number of images in the test folder = 158 images of both charlock and cleavers
• Total number of images in the validation folder = 148 images of both charlock and cleavers

Project Objectives

1. Exploring various image pre-processing techniques to pre-process the dataset
2. Extracting different features such as Histogram of Gradient, RGB values, colour Histogram
3. Training and optimizing different machine and deep learning Models, such as support vector machines (SVMs), random forests, Logistic Regression, K-Nearest Neighbour (KNN),Convolutional Neural Networks(CNN) using the extracted features.
4. Evaluating the performance of the classifiers using appropriate evaluation metrics, such as accuracy, precision, recall, and F1-score.
5. Comparing and contrasting the performance of traditional Machine Learning Classifiers with Deep Learning Model (CNN)