Investigating Input Protocols, Image Analysis, and Machine Learning Methods for an Intelligent Identification System of Fusarium Oxysporum Sp. in Soil Samples

• Main Authors: Coronel, Andrei D, Estuar, Ma. Regina Justina E, De Leon, Marlene M
• Format: text
• Published: Archīum Ateneo 2018
• Subjects: Image analysis; Machine learning; Smart farming; Wireless sensor networks; Mobile microscopy; Agriculture; Computer Sciences; Databases and Information Systems; Fungi; Plant Sciences; Soil Science
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/275; https://link.springer.com/chapter/10.1007/978-3-030-01054-6_26
• Institution: Ateneo De Manila University

The export of Cavendish cultivars making up one third of the Philippine Banana Export Industry is threatened by the rising concern over the increasing presence of Fusarium wilt in plantations. As Cavendish is susceptible to the infection of the fungus Fusarium oxysporum sp. cubense, Tropical Race 4 (Foc TR4), there is a need to develop an early detection mechanism whereby farmers can determine whether the soil is susceptible to the fungi and whether planting Cavendish cultivars in the area should be avoided. Developed in the Philippines, CITAS is a cloud based intelligent total analysis system that uses wireless soil sensor networks and mobile microscopy in determining the presence or absence of Foc TR4. The study implemented a two step approach in the development of an intelligent detection system, specifically using image analysis in magnified soil samples for identification and a machine learning approach for intelligent classification and modeling. The results of the study also served to guide the development of the appropriate soil sampling protocol that would be used for the mobile microscope, as well as the design of the mobile microscope itself. Experiments involve shape detection methods on variable-sized image inputs alongside machine learning techniques such as Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN, a specialized type of ANN), and Support Vector Machines (SVM). The best results of the study with regards to classification accuracy is an 82.23% average after cross-fold validation, produced by ANN on 32×" role="presentation" style="box-sizing: border-box; display: inline-table; line-height: normal; letter-spacing: normal; word-spacing: normal; overflow-wrap: normal; white-space: nowrap; float: none; direction: ltr; max-width: none; max-height: none; min-width: 0px; min-height: 0px; border: 0px; padding: 0px; margin: 0px; position: relative;">××32 pixel image inputs prepared under phase contrast microscopy. However, considering the subsequent implementation of the system on a mobile platform, favored metric is therefore the fastest processing time, yielded by SVM at 8.09 s under similar specimen parameters, with an acceptable accuracy of 81.12%. It is conclusive that a stained soil preparation protocol paired with 100x magnification produces a viable input for a shape-recognition based image analysis technique. The study is an important step towards a multi-parameter approach in the early detection of Foc TR4 infection, thus potentially changing farmer behavior from reactive practices to preventive measures in the context of Fusarium Wilt.