Implementation of MobileNetV2 SSD FPN-Lite CNN Model for Real-time Detection Spinach Leaf Diseases

Abstract

Early detection of disease in spinach plants is critical in preventing the spread of infection and yield losses. The main objective of this research is to develop an application that is effective, fast, and accurate in detecting diseases on spinach leaves by utilizing Convolutional Neural Network (CNN) technology MobileNetV2 SSD FPN-Lite and Tensorflow-Lite models for implementation on the Android platform and knowing the inference time of the model used in detecting diseases. The dataset used in this study is an image of spinach leaves infected with white rust disease, curly virus, and manganese deficiency. The model training process is carried out using the transfer learning method on the dataset that has been created. This research contributes by providing a disease detection application with high accuracy and fast inference time and consumes less computing resources. The results showed that the model created could detect diseases in spinach leaves with an average value of inference and inference timing in testing using 290 images of 12784 (μs) and 39406.2 (μs), and accuracy reached 89.65%. With high accuracy and good computational efficiency, this application can be an effective and efficient solution to support farmers and agricultural experts in the early detection of spinach plant diseases.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint