Deep Learning-Based Classification of Image Data Sets Containing 111 Different Seeds

Image analysis plays a crucial role in understanding and protecting biodiversity. A wide variety of images are used in research on identifying and classifying plants, including stems, leaves, flowers, and fruits. In order to increase crop production, more research needs to be done on the image analysis of seeds. This study aims to fill the gap in this field by creating an image data set of 111 different species in 42 families. An improved Convolutional Neural Networks (CNNs) model is developed by adding new layers to the last layers of the well-known CNNs in the literature. A well-balanced image data set is used to train the proposed model and calculate its performance. The accuracy of the custom CNNs model for seed classification is between 91% and 94%. The custom model's top-2 and top-3 accuracy values are 98.56% and 98.92%, respectively. The proposed CNNs model shows encouraging results in terms of accuracy and computation time for seed classification and recognition. A comprehensive database containing 6536 images of 111 seeds from 42 families is created and is publicly available to scientists for further analysis. A custom CNN model is developed and trained using the database. Different activation functions and batch sizes are used to evaluate the performance of the proposed model. Having achieved convincing results, the seed classification process can now be completely automated by a computer vision system.image