Coconut Pest and Disease
Surveillance System

Agriculture and related industries employed more than a quarter of the Sri Lankan workforce by the year 2019 [1]. However, due to frequent changes in climatic conditions, disease spread has increased resulting in severe crop losses in Sri Lankan cultivation [1]. Disease and pest identification and assessment techniques based on learning features automatically to collect valuable information will be sustainable in the future [2]. According to Sladojevic et. al. neural networks can extract patterns and identify trends that are too complex for humans or computer systems to notice [3]. This has been proved by a study that proposed a Convolutional Neural Network (CNN) based model responsible for detecting leaves and distinguishing between healthy and diseased leaves of 13 diseases with an overall accuracy of 96.3% [3].

The author, Chandy [2] proposed a system for identifying coconut disease and pest infestations using deep learning techniques, in which symptoms are identified by using a drone that will fly through the coconut lands to capture images. It will identify the pest infestation by processing the images using deep learning and image processing. Finally, the result is sent to the user’s mobile phone using WIFI [2].

Proposed Problem

How to classify pest and diseases in coconut and provide surveillance to people in real time?

Weligama Coconut Leaf Wilt Disease (WCLWD) and Coconut Caterpillar Infestation (CCI) are the most threatening disease and pest of coconut in Sri Lanka. Due to WCLWD about 300,000 palms were removed and an estimated 60,000 palms needed to be identified for removal. An economical yield loss will result if more than 30% defoliation is caused due to CCI. For efficient management of WCLWD and CCI, identification at an early stage and effective communication of growers and professionals is needed.

The proposed pest and disease Surveillance system consists of 4 main components. They are;

1. WCLWD and its symptom severity identification
2. CCI Identification and progression level determination
3. Deficiency Identification (Mg deficiency and Leaf Scorch Decline)
4. Water resource identification
5. Crowdsourcing for information sharing.

Fig 4 illustrates the overall system diagram of the proposed solution which was intended to provide a smart approach for stakeholders, researchers, and Coconut Development Officers (CDOs) to detect the coconut diseases and pest infestations that may affect the coconut industry. As shown in the diagram, the registered users of the system can capture or upload the images which are suspicious. The images are sent to the Amazon Web Services (AWS) backend server where the flask server is deployed. These images are processed in the flask server by the designed DCNN models for disease identification. If WCLWD is found, the symptom severity will be determined using CNN models. Simultaneously, if CCI is identified, the images will go through the Mask-R-CNN model to determine the progression level while the number of caterpillars are extracted using the YOLOv5 object detection algorithm. Images will be classified using the CNN models of Mg deficiency and LSD at the same time.

Once the system identifies that the leaves are infected, then the response will be captured by the crowdsourcing platform. The Google Map will be updated with the real time locations (latitude and longitude) of the infected palms. In Addition, the system will automatically send notifications to the farmers and other stakeholders who are at the risk of infection.