Efficient Disease and Pests Detection in Cotton Leaves Using NASBiT with Enhanced XAI visualization

Abstract

Cotton is amongst the most important crops and diseases and pests have large impacts on them, hence, they have large impacts on agriculture and therefore the global economy. It has been established that the identification of these challenges is crucial in a bid to maintain the crop's health and boost yields. The study proposes NASBiT as a novel hybrid model for accurately indetifying diseases and pests on cotton leaves. NASNet-Mobile combined with Big Transfer (BiT-M-R50x1) focuses on pattern recognition and management of visual features, which makes NASBiT outstanding. Through the fusion of NASNet-mobile and Big Transfer (BiT-M-R50x1), NASBiT demonstrates exceptional aptitude in extracting patterns and managing visual features adeptly. NASBiT exhibits noteworthy metrics including a precision of 100%, recall of 99.83%, F1 score of 99.91%, and an accuracy rate of 98.98%, surpassing existing models in reliability and precision highlighting its commendable performance. Besides, it enhances understanding of the advantages of using hybrid models in agricultural environments and their ability to reshape the detection of diseases in the leaves of cotton crops. However, some limitations include the need for comprehensive empirical for verification across diverse environmental conditions, and research into additional properties that can enhance the facility's performance even more, though briefly.