Authors

Using the YOLOv3 algorithm with preand post-processing procedures for fruit detection by an apple-picking robot

The development of robotic harvesting can help reduce the share of heavy manual labor in horticulture that reaches now 40%, as well as crop shortages, which reach up to 50%. Fruit picking robots have been developing since the late 1960s. However, no existing prototype is used in practice due to the low speed of harvesting and the large proportion of unrecognized fruits remaining on trees. The paper aims to develop an algorithm for detecting apples in images that can work quickly and find as many apples as possible. For this purpose, the use of the YOLOv3 convolutional neural network has been proposed, accompanied by special pre- and post-processing procedures. The procedures aim to improve the quality of apple recognition, including situations of the presence of shadows, glare, various damages to apples, empty gaps between the leaves, which could be mistaken for apples, overlapping apples by branches, leaves, and other apples. The algorithm recognizes both red and green apples. It can work with images of single apples in close-up photographs as well as with images of many apples in general pictures. The algorithm quality was evaluated on a test set of 818 images of red and green apples (5142 apples in total). The average apple detection time was 19 ms, the percentage of objects mistaken for apples turned out to be at the level of 7.8%, and the share of undetected apples at 9.2%. Both the average detection time and the error rates turned out to be noticeably shorter than in all known similar systems. Read more...

Statistical Precision – Recall curves for object detection quality assessment

Average precision (AP) as the area under the Precision – Recall curve is the de facto standard for comparing the quality of algorithms for classification, information retrieval, object detection, etc. However, traditional Precision – Recall curves usually have a zigzag shape, which makes it difficult to calculate the average precision and to compare algorithms. This paper proposes a statistical approach to the construction of Precision – Recall curves when assessing the quality of algorithms for object detection in images. This approach is based on calculating Statistical Precision and Statistical Recall. Instead of the traditional confidence level, a statistical confidence level is calculated for each image as a percentage of objects detected. For each threshold value of the statistical confidence level, the total number of correctly detected objects (Integral TP) and the total number of background objects mistakenly assigned by the algorithm to one of the classes (Integral FP) are calculated for each image. Next, the values of Precision and Recall are calculated. Precision – Recall statistical curves, unlike traditional curves, are guaranteed to be monotonically non-increasing. At the same time, the Statistical Average Precision of object detection algorithms on small test datasets turns out to be less than the traditional Average Precision. On relatively large test image datasets, these differences are smoothed out. The comparison of the use of conventional and statistical Precision – Recall curves is given on a specific example. Read more...