Simplified Image Classification for Nigeria's Agricultural Produce through Deep Neural Network Techniques
Authors: Oyebode KO, Osah UJ
DOI Info: http://doi.org/10.5281/zenodo.6725124
Image processing techniques can support quality checks for agricultural produce. Noting the relevance to the Nigerian context, this paper details the development of a computer vision system for automatic screening of produce. The development included four phases, namely: model development, training, graphical user interface (GUI) creation, and module testing. Model development involved a convolution process with capacity to extract useful features from the image of an agricultural produce. Training enabled the created model to learn crucial image structures based on fine-tuned mask parameters that support the classification of similar images efficiently. The GUI enables non-technical users to train new models, as well as carry out single and multiple classifications. The testing phase enables the evaluation of the system. Images are tested using a pre-trained model as well as without a pre-trained model. The model backed by the pre-trained model performed better than that not associated with a pre-trained model. Importantly, whereas the application of deep neural networks has been explored, a useful contribution here is the introduction of a user friendly GUI to underpin its employment.
Affiliations: Department of Design and New Media, School of Media and Communication, Pan-Atlantic University, PMB 73688, Lagos, Lagos State, Nigeria.
Keywords: Deep Neural Network, Image Processing, Convolution, Computer Vision, Quality Checks, Agriculture
Published date: 2022/06/30