Prediction of diseased rice plant using video processing and LSTM-simple recurrent neural network with comparative study

摘要

The disease infliction of the plants severely influences the yield. It alters the essence and extent of crop production cause fiscal distress. Consequently, the diagnosis of numerous plant diseases is significant to decrease the yield perdition by discovering crop infections at their earlier stages. This paper introduces a new model using mobile video image processing and Long-Short Term Memory (LSTM)-Simple Recurrent Neural Network (SRNN) deep learning method for the prediction of the diseased or disinfected rice plant with dynamic learning capability. The rice plant videos captured under uncontrolled conditions in day-lighting using a mobile handset and divided into two sections for the designing and testing of LSTM-SRNN models. After shooting, the video images of the rice plant segmented using colour indexing and linear color space transformation with minimal daylight impact. Low-level spatial features; entropy, standard deviation, and fuzzy features extracted after video image segmentation. The excerpted characteristics with the composite combinations transformed in time-series datasets with the desired response. The datasets employed in the LSTM-SRNN model for progressive learning. The distinct test video features applied in LSTM-SRNN to appraise the generalization capability of the proposed system with performance analysis. The experimental outcomes of the proposed LSTM-SRNN model exhibit 99.99% prediction ability with fuzzy features. The model also presents possibilities for dynamic learning adaptability and temporal information processing to overcome the limitations of many well-known rule-based and machine learning approaches.