Blueberries have high nutritional and economic value and are easy to cultivate, so they are common fruit crops in China. There is a high demand for blueberry in domestic and foreign markets, and various technologies have been used to extend the supply cycle of blueberry to about 7 months. However, blueberry grows in clusters, and a cluster of fruits generally contains fruits of different degrees of maturity, which leads to low efficiency in manually picking mature fruits, and at the same time wastes a lot of manpower and material resources. Therefore, in order to improve picking efficiency, it is necessary to adopt an automated harvesting mode. However, an accurate maturity detection model can provide a prerequisite for automated harvesting technology. Therefore, this paper proposes a blueberry ripeness detection model based on enhanced detail feature and content-aware reassembly. First of all, this paper designs an EDFM (Enhanced Detail Feature Module) that improves the ability of detail feature extraction so that the model focuses on important features such as blueberry color and texture, which improves the model’s ability to extract blueberry features. Second, by adding the RFB (Receptive Field Block) module to the model, the lack of the model in terms of receptive field can be improved, and the calculation amount of the model can be reduced at the same time. Then, by using the Space-to-depth operation to redesign the MP (MaxPool) module, a new MP-S (MaxPool–Space to depth) module is obtained, which can effectively learn more feature information. Finally, an efficient upsampling method, the CARAFE (Content-Aware Reassembly of Features) module, is used, which can aggregate contextual information within a larger receptive field to improve the detection performance of the model. In order to verify the effectiveness of the method proposed in this paper, experiments were carried out on the self-made dataset “Blueberry—Five Datasets” which consists of data on five different maturity levels of blueberry with a total of 10,000 images. Experimental results show that the mAP (mean average precision) of the proposed network reaches 80.7%, which is 3.2% higher than that of the original network, and has better performance than other existing target detection network models. The proposed model can meet the needs of automatic blueberry picking.

The damage caused by pests to crops results in reduced crop yield and compromised quality. Accurate and timely pest detection plays a crucial role in helping farmers to defend against and control pests. In this paper, a novel crop pest detection model named YOLOv5s-pest is proposed. Firstly, we design a hybrid spatial pyramid pooling fast (HSPPF) module, which enhances the model’s capability to capture multi-scale receptive field information. Secondly, we design a new convolutional block attention module (NCBAM) that highlights key features, suppresses redundant features, and improves detection precision. Thirdly, the recursive gated convolution ($g^{3}Conv$) is introduced into the neck, which extends the potential of self-attention mechanism to explore feature representation to arbitrary-order space, enhances model capacity and detection capability. Finally, we replace the non-maximum suppression (NMS) in the post-processing part with Soft-NMS, which improves the missed problem of detection in crowded and dense scenes. The experimental results show that the mAP@0.5 (mean average precision at intersection over union (IoU) threshold of 0.5) of YOLOv5s-pest achieves 92.5% and the mAP@0.5:0.95 (mean average precision from IoU 0.5 to 0.95) achieves 72.6% on the IP16. Furthermore, we also validate our proposed method on other datasets, and the outcomes indicate that YOLOv5s-pest is also effective in other detection tasks.

As one of the most important economic crops, grapes have attracted considerable attention due to their high yield, rich nutritional value, and various health benefits. Identifying grape bunches is crucial for maintaining the quality and quantity of grapes, as well as managing pests and diseases. In recent years, the combination of automated equipment with object detection technology has been instrumental in achieving this. However, existing lightweight object detection algorithms often sacrifice detection precision for processing speed, which may pose obstacles in practical applications. Therefore, this thesis proposes a lightweight detection method named YOLOv8s-grape, which incorporates several effective improvement points, including modified efficient channel attention (MECA), slim-neck, new spatial pyramid pooling fast (NSPPF), dynamic upsampler (DySample), and intersection over union with minimum point distance (MPDIoU). In the proposed method, MECA and NSPPF enhance the feature extraction capability of the backbone, enabling it to better capture crucial information. Slim-neck reduces redundant features, lowers computational complexity, and effectively reuses shallow features to obtain more detailed information, further improving detection precision. DySample achieves excellent performance while maintaining lower computational costs, thus demonstrating high practicality and rapid detection capability. MPDIoU enhances detection precision through faster convergence and more precise regression results. Experimental results show that compared to other methods, this approach performs better in the grapevine bunch detection dataset and grapevine bunch condition detection dataset, with mean average precision (mAP50–95) increasing by 2.4% and 2.6% compared to YOLOv8s, respectively. Meanwhile, the computational complexity and parameters of the method are also reduced, with a decrease of 2.3 Giga floating-point operations per second and 1.5 million parameters. Therefore, it can be concluded that the proposed method, which integrates these improvements, achieves lightweight and high-precision detection, demonstrating its effectiveness in identifying grape bunches and assessing biophysical anomalies.

Estimating accurate 3D hand pose from a single RGB image is a highly challenging problem in pose estimation due to self-geometric ambiguities, self-occlusions, and the absence of depth information. To this end, a novel Five-Layer Ensemble CNN (5LENet) is proposed based on hierarchical thinking, which is designed to decompose the hand pose estimation task into five single-finger pose estimation sub-tasks. Then, the sub-task estimation results are fused to estimate full 3D hand pose. The hierarchical method is of great benefit to extract deeper and better finger feature information, which can effectively improve the estimation accuracy of 3D hand pose. In addition, we also build a hand model with the center of the palm (represented as Palm) connected to the middle finger according to the topological structure of hand, which can further boost the performance of 3D hand pose estimation. Additionally, extensive quantitative and qualitative results on two public datasets demonstrate the effectiveness of 5LENet, yielding new state-of-the-art 3D estimation accuracy, which is superior to most advanced estimation methods.

3D hand pose estimation can provide basic information about gestures, which has an important significance in the fields of Human-Machine Interaction (HMI) and Virtual Reality (VR). In recent years, 3D hand pose estimation from a single depth image has made great research achievements due to the development of depth cameras. However, 3D hand pose estimation from a single RGB image is still a highly challenging problem. In this work, we propose a novel four-stage cascaded hierarchical CNN (4CHNet), which leverages hierarchical network to decompose hand pose estimation into finger pose estimation and palm pose estimation, extracts separately finger features and palm features, and finally fuses them to estimate 3D hand pose. Compared with direct estimation methods, the hand feature information extracted by the hierarchical network is more representative. Furthermore, concatenating various stages of the network for end-to-end training can make each stage mutually beneficial and progress. The experimental results on two public datasets demonstrate that our 4CHNet can significantly improve the accuracy of 3D hand pose estimation from a single RGB image.