Article Preview
Top1. Introduction
Soybean, as one of the important grain and oil crop in the world, plays an important role in the world's agricultural production and trade (Wu, Zhang, & Meng, 2019). However, various soybean diseases have constrained sustainable development of high-yield and high-quality soybean industry for a long time. For one thing, there will likely be a large increase in the demand for soybean with the growth of population and economy. For another thing, soybean diseases have characteristics of large variety, great impact and local outbreaks, which have been responsible for productivity and quantitative losses in crop yield. Thus, time-saving and high-efficiency identification of soybean diseases is urgently needed.
Classification algorithms play a substantial role in crop diseases identification. With the development of sophisticated instruments and fast computational techniques, the application of machine learning technologies to diagnose crop diseases has become one of the important research contents of intelligent agriculture. Although advances in science and technology now make it possible for computer vision approaches to assist us in automatic detection crop diseases tasks and remarkable performances have been achieved, these classification methods based on deep learning are limited to using large amount of high-quality image data and depending on great computational capacity. Our research is devoted to using simple and efficient classification method and few diseased samples for soybean diseases identification task. Symptom features can be conveyed by plant organs such as seeds, leaves and fruits, which are generally sensitive to the state of crops and whose shape, texture and color usually contain rich information. And thus the identification of soybean diseases can be based on descriptive data. Compared with image-based methods, the statistical pathological inputs can also achieve high recognition accuracy but fast evaluation speed.
ANNs use supervised learning to determine a complex, nonlinear, multidimensional mathematical fitting and have attracted a great deal of attention recently. ANNs have so excellent generalization ability and robustness that they excel in many areas: pattern classification, function approximation, intelligent control, fault detection, signal processing and system analysis etc. In addition, ANNs have also achieved impressive results in the field of agriculture and benefit more smallholders and horticultural workers, including weeds recognition (Ahmad et al., 2018), plant diseases prediction (Sharma, Singh, & Singh, 2018), soil parameters estimation (Estrada-López, Castillo-Atoche, Vázquez-Castillo, & Sánchez-Sinencio, 2018) and land cover detection (Zhan, Tian, & Tian, 2019) etc. This paper presents an application of PSO-based multi-layer perceptron (MLP) neural network in soybean diseases identification. The proposed method is capable of providing a reliable and fast estimation of soybean diseases types on the basis of typical symptom features. The performance of the proposed technique is analyzed by comparing the classification results with the traditional machine learning methods including logistic regression (LR), k-NearestNeighbor (KNN) and support vector machine (SVM) for the same hold-out test dataset.