Dynamic Voting-Based Ensemble Deep Learning for Closely Resembling Crop Classification

İngilizce (OpenAlex)

Utilizing modern deep learning techniques for image processing and data classification holds immense promise for yielding significant outcomes. Consequently, deep learning-based approaches have demonstrated successful applications in agricultural contexts, particularly in tasks such as image classification and data analysis. This study focuses on classifying agricultural crop images, which often bear close resemblance, employing 17 distinct deep learning models and dynamic voting. Initially, the paper provides a concise overview of the dataset and the deep learning methodologies employed. Subsequently, the training and testing phases are carried out effectively. The dataset comprises a total of 804 images depicting five different types of crops: jute, maize, rice, sugarcane, and wheat. To ensure robustness, 10-fold cross-validation is employed, and experiments are conducted consistently across all models using the same sample sets. The results obtained report which models can more accurately detect and classify agricultural crop images. Further, the proposed ensemble approach improves accuracy and ensures greater robustness and stability. According to the experimental findings, Shufflenet achieved the highest individual accuracy of 98.63% on the test set, but the ensemble approach increased this value to 99.75%.