A Recognition Method for Cassava Phytoplasma Disease (CPD) Real-Time Detection based on Transfer Learning Neural Networks

Abstract

Object detection technology aims to detect the target objects with the theories and methods of image processing and pattern recognition, determine the semantic categories of these objects, and mark the specific position of the target object in the image. This study generally aims to establish a recognition method for Cassava Phytoplasma Disease (CPD) real-time detection based on transfer learning neural networks. Several methods and procedures were conducted, such as the testing of two methods in transmitting long-distance high definition (HD) video capture; establishment of a compact setup for a long-range wireless video transmission system; the development, testing of the real-time CPD detection and quantification monitoring system, providing the comparative performance analysis of the three models used. We have successfully custom-trained three artificial neural networks using transfer learning: Faster Regions with Convolutional Neural Networks (R-CNN) Inception v2, Single Shot Detector (SSD) Mobilenet v2, and You Only Look Once (YOLO) v4. These deep learning models can detect and recognize CPD in actual environment settings. Overall, the developed real-time CPD detection and quantification monitoring system was successfully integrated into the wireless video receiver and seamlessly visualized all the incoming data using the three different CNN models. If the consideration is the image processing speed, YOLOv4 is better compared to other models. But, if accuracy is the priority, Faster R-CNN inception v2 performs better. However, since CPD detection is the main purpose of this study, the Faster R-CNN model is recommended for adoption to detect CPD in a real-time environment.