Machine learning ensemble with image processing for pest identification and classification in field crops

In agriculture field, yield loss is a major problem due to attack of various insects in field crops. Traditional insect identification and classification methods are time-consuming and require entomologist experts. Early information about the attack of insects helps farmers to control the crop damage to improve the productivity and reduce the use of pesticides. This research work focuses on the classification of crop insects by applying machine vision and knowledge-based techniques with image processing by using different feature descriptors including texture, color, shape, histogram of oriented gradients (HOG) and global image descriptor (GIST). A combination of all these features was used in the classification of insects. In this research, several machine learning algorithms including both base classifiers and ensemble classifiers were applied for three different insect datasets and the performances of classification results were evaluated by majority voting. Naive bayes (NB), support vector machine (SVM), K-nearest-neighbor (KNN) and multi-layer perceptron (MLP) were used as base classifiers. Ensemble classifiers include random forest (RF), bagging and XGBoost were utilized; 10-fold cross-validation test was conducted to achieve a better classification and identification of insects. The experimental results showed that the classification accuracy is improved by majority voting with ensemble classifiers in the combination of texture, color, shape, HOG and GIST features.