Master Thesis - Computer Vision

<p>Identifying optimal camera setup that allows machine vision application for plant part detection, counting, and measurement. </p><p></p><p>Background and Context</p><p>Wheat is one of the most important cereal crops globally, serving as a staple food source for millions of people. Accurate wheat plant part detection, counting, and measurements play a crucial role in assessing crop health, predicting yields, and optimizing agricultural practices. Traditional manual counting/measurement methods are time-consuming and prone to human error. Leveraging modern technology, such as computer vision and machine learning, can significantly improve the accuracy and efficiency of this task.</p><p></p><p>This master's thesis project aims to address the critical need for automated wheat 2D/3D plant part detection, counting, and measuring contributing to the advancement of precision agriculture. By developing cutting-edge machine learning algorithms and identifying the optimal camera setup, this project seeks to enhance crop monitoring and management, ultimately benefiting sustainable agricultural production and food security.</p><p></p><p>Objectives:</p><p>Identifying the best camera setup for wheat plant part detection:</p><p>This objective entails evaluating various camera configurations, including camera types, resolutions, angles, and lighting conditions, to determine the optimal setup for capturing wheat heads, spikelets, leaf angles, stem width etc. in a greenhouse or field conditions. The goal is to identify the most economical and suitable camera sensor setup that maximizes image quality, minimizes noise, and ensures consistent data acquisition under changing environmental conditions. Different camera sensors such as RGB, multi or hyperspectral, fluorescence, and thermal cameras will be used to extract information about plant health. Adaptation of existing object detection algorithms and/or development of new algorithms will be needed. </p><p></p><p>Achieving the Objectives</p><p>To achieve the proposed objectives of identifying the best camera setup, the following steps and methodologies will be employed</p><p></p><p>Data Collection: Collect a diverse dataset of wheat plant images using various camera configurations, 2D and 3D images, capturing different growth stages.</p><p></p><p>Data Preprocessing: Clean and preprocess the collected images to enhance their quality and consistency.</p><p></p><p>Experimental Design: Conduct systematic experiments to assess the performance of different camera setups, measuring key parameters such as image resolution, focal length, and lighting conditions.</p><p></p><p>Evaluation Metrics: Use quantitative metrics like image quality scores and detection accuracy to evaluate camera setups.</p><p></p><p>Selecting the Optimal Setup: Analyze the experimental results and choose the camera setup that consistently produces high-quality images for the wheat plant part detection and health prediction.</p><p></p><p>Previous work:</p><p>Camera setup:</p><p><a href="https://phenospex.com/blog/an-overview-of-3d-plant-phenotyping-methods/" rel="noopener noreferrer" target="_blank">https://phenospex.com/blog/an-overview-of-3d-plant-phenotyping-methods/</a></p><p><a href="https://www.frontiersin.org/articles/10.3389/fpls.2022.897746/full" rel="noopener noreferrer" target="_blank">https://www.frontiersin.org/articles/10.3389/fpls.2022.897746/full</a></p><p><a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7040167/" rel="noopener noreferrer" target="_blank">https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7040167/</a></p><p></p><p>AI (for letter stage)</p><p><a href="https://www.frontiersin.org/articles/10.3389/fpls.2022.872555/full" rel="noopener noreferrer" target="_blank">https://www.frontiersin.org/articles/10.3389/fpls.2022.872555/full</a></p><p><a href="https://www.mdpi.com/2077-0472/13/4/872" rel="noopener noreferrer" target="_blank">https://www.mdpi.com/2077-0472/13/4/872</a></p><p><a href="https://ueaeprints.uea.ac.uk/id/eprint/71173/1/SpikeletFCN.pdf" rel="noopener noreferrer" target="_blank">https://ueaeprints.uea.ac.uk/id/eprint/71173/1/SpikeletFCN.pdf</a></p><p></p><p>Contact persons:</p><p></p><p>Mikael Nilsson: <a href="mailto:mikael.nilsson@math.lth.se" rel="noopener noreferrer" target="_blank">mikael.nilsson@math.lth.se</a></p><p>Marc Ahlse : <a href="mailto:Marc.Ahlse@sony.com" rel="noopener noreferrer" target="_blank">Marc.Ahlse@sony.com</a></p><p>Ajit Nehe: <a href="mailto:ajit.nehe@slu.se" rel="noopener noreferrer" target="_blank">ajit.nehe@slu.se</a></p>