Crop health analysis system: Integrating machine learning for disease detection in agricultural images

Abstract

The proposed project is a pioneering research effort utilizing cutting edge technology to revolutionize plant disease detection within agriculture. Utilizing CNN for feature extraction and ResNet-50 for classification, this system outperforms in the accuracy of providing crop disease from the agricultural images. This approach utilizes a formal Image Classification Cycle, which offers processes for data acquisition, augmentation оr data processing, thereby improving the accuracy and efficiency of detection. Novel features entail merging sophisticated machine learning models with a simple and clean web application via the Streamlit framework allowing on-line disease identifying for farmers. When comparing the accuracy and performance of the system against the leading ResNet-50 algorithm, it is noted that the accuracy is 95% with optimized ResNet-50 achieving 98.72% accuracy. This implementation combines the speed, accuracy, and scalability of manual observation and traditional algorithms. It also discusses functional requirements, such as dataset reliability, and non-functional properties, like usability, robustness, and adaptability to a diverse agricultural condition. Finally, this work is distinguished from other solutions by its speed, scalability for small and large farms, and flexibility to environmental challenges. The system, which can be used by farmers as a tool, is expected to increase crop yield, minimize dependence on pesticides, and promote sustainable agricultural practices. The results highlight the capacity of machine learning powered systems to change agriculture and support global food security.