Smart Agriculture: CNN-RF Model for Automated Rice Blast Severity Detection

An efficient model based on deep learning for defect classification presented in this paper can classify and delineate defects according to several scales. The model was found to show Recall, Precision, F1-Score of around 98.45% to 98.62%, clearing the way for less misclassification and good defect detection. The Support Proportion values in the interval of 0.19-0.21 confirm a fairly well-distributed dataset that fully covers all the defect scales. Also, a perfect accuracy of 99% for all defect classes is another evidence of the reliability and the stability of this model in defect classification. The evaluation metrics also substantiate the fact that the model is successful in defect detection. These high True Positive count and low False Positive (30) and False Negative (30) values are suggestions of the high classification accuracy in practically minimizing errors. A large database comprising 10,280 images was used to train and test the model-ingrained strength and robustness toward different types of defects. Thus, with an overall accuracy attaining up to 98.54%, it finally can be labeled as a reliable and efficient tool for computer-based defect classification in the industry at large. The approach will minimize manual inspection but increase accuracy and consistency. This indicates the potential that deep learning technology holds for industrial quality control, with solutions that are scalable and efficient. Future work may consider adding more classes of defects and tuning model parameters for performance enhancement.