Detecting tomatoes in greenhouse scenes by combining AdaBoost classifier and colour analysis

• Ripe tomatoes in greenhouse detected using AdaBoost classifier and colour analysis.
• AdaBoost classifier obtained through “WeakLearn” and Haar-like feature extraction.
• C type of Haar-like feature optimum feature for detecting ripe tomatoes.
• Average pixel value based colour analysis of I component eliminated false positives.
• Proposed detection algorithm tested with the tomato images acquired in greenhouse.

Despite the rapid development of agricultural robotics, a lack of access to automatic fruit detection and precision picking is limiting the commercial application of harvesting robots. An algorithm for the automatic detection of ripe tomatoes in greenhouse was developed for a simple machine vision system. The images of tomato planting scenes were captured by a colour digital camera, and most of the ripe tomatoes were correctly recognised using the proposed algorithm. The proposed tomato detection approach worked in two steps: (1) by extracting the Haar-like features of grey scale image and classifying with the AdaBoost classifier, the possible tomato objects were identified; (2) the false negatives in the results of classification were eliminated using average pixel value (APV) based colour analysis approach. Comparative test results showed that the C style of Haar-like features and I component image were optimum in the proposed algorithm. The results of validation experiments show that combination of AdaBoost classification and colour analysis can correctly detect over 96% of ripe tomatoes in the real-world environment. However, the false negative rate was about 10% and 3.5% of the tomatoes were not detected.