Enhancing Plant Disease Detection Through Image Analysis Using SSDmobilenetV2 and ResNet50

Being an agrarian nation, Nepal possesses huge economic value from the agricultural sectors with more than half of its population contribution in farming, thus bringing a great potential share in the nation's Gross Domestic Product (GDP).Despite this substantial contribution, there are still issues with advancement that hinders growth in this field.This study, therefore presents a model for the detection of plants and diagnosis of diseases associated with the plants using real-time camera feeds or by analyzing captured images.Existing models often suffer from unrelated irrelevant images; confusing one plant for another which results in failure of proper diagnosis of diseases.Addressing this issue, our model first detects the presence of a plant within the frame or image before disease classification.Plant detection is performed using the Single-Shot Detector (SSD) Mobilenet V2 model.Disease classification process is initiated only when the plant is detected.The Residual Network (ResNet)-50 model performs the plant disease classification taking the clipped image from detected plant.By focusing only on the detected plant, we reduce background complexity and improve classification accuracy.The model has demonstrated a high level of accuracy in both detecting plants and classifying diseases based on the prevalent diseases associated with specific plants.Our model achieved a mean Average Precision (mAP) score of 0.61 for the object detection model which includes 9 classes and achieved average accuracy of 98% for the classification of 9 plant species having a total of 33 classes.This innovative model is hence targeted at providing reliable and accurate plant detection and disease diagnosis to address some of the key challenges in agricultural technology in Nepal.On near future it has the potential for automated plant diseases detection and real time monitoring of plant diseases.Additionally, we can deploy it in mobile applications, allowing farmers to identify plant diseases using smartphone cameras, facilitating timely detection and reduction of crop losses.