Real-Time Crop Health Assessment with MobileNet for Agricultural Applications

The growth of sustainable and efficient farming systems has necessitated the use of automated crop health monitoring systems. In the current study, a real-time crop health assessment is proposed using a deep learning algorithm called MobileNet based on a lightweight CNN. The method under consideration includes using the camera of a smartphone to take detailed photographs of crops growing in the fields, and with the help of the MobileNet model, distinguish diseases, nutrient deficiencies, and other pathologies. To provide some resistance to variations in environmental circumstances, the current images are preprocessed through resizing, normalization, and augmentation. MobileNet model fine-tuning is made using a richer dataset that includes both healthy and disease crop images The model achieves an accuracy of 94.5% and micro-average of precision, recall, and F1-score of 95.3%, 93.7%, and 94.5% respectively. Real-time efficiency tests are performed on disparate mobile devices and show that the system took less than 2 seconds on average to generate results and had a model size of 12 MB, suitable for usage on even minimalresource devices. Moreover, the trained model demonstrates relatively irgendeepstance to image brightness and orientation, which is rather advantageous for its practical application on the agricultural fields. This paper presents a possible practical and effective approach to field use, which will allow for improving precision agriculture, as well as the preservation of the environment and agricultural lands.