A Dense Litchi Target Recognition Algorithm for Large Scenes

Abstract

To address the automatic detection of dense and small-scale fruit targets under natural large-scene conditions, litchi was used as the research object. Here, a method to automatically detect dense and small-scale litchi fruit targets based on the YOLOv4 detection network is proposed. First, the K-means++ algorithm was used to cluster the labelled data frames (ground truth) to determine the size of the anchor suitable for litchi. Then, the output size of the feature map of the original network was changed to make it more suitable for small-scale target detection. In addition, the images were preprocessed (cropped input) before they were fed into the network. To construct the litchi dataset, 400 images containing more than 20,000 targets were collected. Comparing the detection level to that of the original YOLOv4 model, the recall, precision, and F1 score values of the improved model increased from 0.81 to 0.825, 0.762 to 0.892, and 0.79 to 0.85, respectively. The experimental results indicate that the performance of the litchi detection method proposed in the study is significantly greater than the original model, and it meets the requirements for fruit monitoring in litchi orchards.

An Overview of the Application of Machine Vision in Recognition and Localization of Fruit and Vegetable Harvesting Robots

“…Based on YOLOv3, the feature extraction capability of the model can be enhanced by performing K-means clustering analysis on the predicted bounding boxes of leafy objects [43,92]. South China Agricultural University [93] proposed a small-scale lychee fruit detection method based on YOLOv4. This method utilizes the K-means++ algorithm to cluster the labeled frames to determine anchor sizes suitable for lychees.…”

An Improved Rotating Box Detection Model for Litchi Detection in Natural Dense Orchards

“…Regarding litchi detection research based on deep learning, Peng et al [25] proposed a novel network model that used the feature pyramid to retain the shallow features of litchis and cut down the model size. Wu et al [7] proposed a litchi detection algorithm based on the YOLOv4 network, using the K-means++ algorithm to select the appropriate size for litchis, and changed the feature map to account for the small and dense litchi targets, but the improved algorithm still missed detection in large scenes. Furthermore, a long-close distance coordination control strategy was proposed based on an RGB-depth camera combined with a point cloud map.…”

“…In actual production, orchard managers usually use random sampling and manual counting to estimate orchard yields [5]. However, scattered litchi clusters, obstructions due to leaves and branches, and the complex lighting conditions of the natural environment lead to low-accuracy litchi identification [6,7]. Consequently, achieving real-time and accurate litchi identification in natural dense scenes is one of the core issues in improving the accuracy of litchi yield estimations.…”