Abstract
The importance of agriculture and the production of fruits and vegetables has stood out mainly over the past few years, especially for the benefits for our health. In 2021, in the international year of fruit and vegetables, it is important to encourage innovation and evolution in this area, with the needs surrounding the different processes of the different cultures. This paper compares the performance between two datasets for robotics fruit harvesting using four deep learning object detection models: YOLOv4, SSD ResNet 50, SSD Inception v2, SSD MobileNet v2. This work aims to benchmark the Open Images Dataset v6 (OIDv6) against an acquired dataset inside a tomatoes greenhouse for tomato detection in agricultural environments, using a test dataset with acquired non augmented images. The results highlight the benefit of using self-acquired datasets for the detection of tomatoes because the state-of-the-art datasets, as OIDv6, lack some relevant characteristics of the fruits in the agricultural environment, as the shape and the color. Detections in greenhouses environments differ greatly from the data inside the OIDv6, which has fewer annotations per image and the tomato is generally riped (reddish). Standing out in the use of our tomato dataset, YOLOv4 stood out with a precision of 91%. The tomato dataset was augmented and is publicly available (See https://rdm.inesctec.pt/ and https://rdm.inesctec.pt/dataset/ii-2021-001).

Abstract
The harvesting operation is a recurring task in the production of any crop, thus making it an excellent candidate for automation. In protected horticulture, one of the crops with high added value is tomatoes. However, its robotic harvesting is still far from maturity. That said, the development of an accurate fruit detection system is a crucial step towards achieving fully automated robotic harvesting. Deep Learning (DL) and detection frameworks like Single Shot MultiBox Detector (SSD) or You Only Look Once (YOLO) are more robust and accurate alternatives with better response to highly complex scenarios. The use of DL can be easily used to detect tomatoes, but when their classification is intended, the task becomes harsh, demanding a huge amount of data. Therefore, this paper proposes the use of DL models (SSD MobileNet v2 and YOLOv4) to efficiently detect the tomatoes and compare those systems with a proposed histogram-based HSV colour space model to classify each tomato and determine its ripening stage, through two image datasets acquired. Regarding detection, both models obtained promising results, with the YOLOv4 model standing out with an F1-Score of 85.81%. For classification task the YOLOv4 was again the best model with an Macro F1-Score of 74.16%. The HSV colour space model outperformed the SSD MobileNet v2 model, obtaining results similar to the YOLOv4 model, with a Balanced Accuracy of 68.10%.

Abstract