Classification of temporal images

Samet Yıldızeli

Master's thesis

Abstract:

The classification of temporal images is a crucial task in the field of computer vision and has numerous applications, including seed germination detection. In this thesis, two models were developed and trained for the classification of temporal images, with a focus on detecting seed germination in soil. The first model, YOLO, was trained for the detection of the first appearance of the plant on the soil. The second model, ConvLSTM, was trained for the detection of the germination of the seed in soil using temporal image classification. The YOLO model was trained using a customized germination dataset containing two classes: "soil" and "FA", which represented the first appearance of the plant on the soil. The results of the training process showed that both models achieved high levels of accuracy in detecting the respective classes. The YOLO model achieved precision values of 0.989 and 0.991 for the "soil" and "FA" classes, respectively, and a mean average precision of 0.995 for both classes. The ConvLSTM model achieved an accuracy of 0.9378, with a loss of 0.2469, demonstrating its effectiveness in detecting seed germination in soil using temporal image classification. This thesis presents the development and training of two models for the classification of temporal images, with a focus on detecting seed germination in soil. The results show that both models are effective in detecting the respective classes, and the ConvLSTM model is particularly well suited for detecting seed germination in soil using temporal image classification.

Keywords

Temporal Image Classification, Seed Germination Detection, YOLO, ConvLSTM, Computer Vision, Image Processing

Language used: English

Date on which the thesis was submitted / produced: 3. 4. 2023

Thesis defence

• Date of defence: 24. 5. 2023
• Supervisor: Ing. Jan Masner, Ph.D.
• Reader: Ladislav Benda, externi

Czech University of Life Sciences Prague

Faculty of Economics and Management

Master programme:
Informatics