A Detection Method for Tomato Fruit Common Physiological Diseases Based on YOLOv2

Abstract

The detection of tomato fruit physiological diseases can increase tomato yields and facilitate tomato fruit quality control. In this paper, a convolutional neural network called YOLOv2 was applied on the detection of healthy tomato fruits and tomato fruits with common physiological diseases. YOLOv2 is a target detection algorithm based on regression model, which has fast detection speed and good accuracy. In order to improve the network detection performance, the following improvements were made: First, image datasets were extended by data augmentation methods to decrease the chance of overfitting, and the augmented datasets containing 1000 tomato fruit images were obtained. Second, a grayscale processing module and a foreground extraction module were constructed to investigate the importance of image data type. Third, k-means clustering algorithm was applied in order to reduce model training time as well as improve detection effects. The mAP (mean Average Precision) of YOLOv2 tomato fruit detection network was 97.24%. Experimental results showed that the proposed method was effective on the detection of healthy tomato fruits and tomato fruits with common physiological diseases.