A Robust Mature Tomato Detection in Greenhouse Scenes Using Machine Learning and Color Analysis

Abstract

A new algorithm for automatic tomato detection in regular color images is proposed, which can reduce the influence of illumination, color similarity as well as suppress the effect of occlusion. The method uses a Support Vector Machine (SVM) with Histograms of Oriented Gradients (HOG) to detect the tomatoes, followed by a color analysis method for false positive removal. And the Non-Maximum Suppression Method (NMS) is employed to merge the detection results. Finally, a total of 144 images were used for the experiment. The results showed that the recall and precision of the classifier were 96.67% and 98.64% on the test set. Compared with other methods developed in recent years, the proposed algorithm shows substantial improvement for tomato detection.