Project Articles

Default Latest Most Read

Please wait a minute... For Selected: Download Citations EndNote Ris BibTeX Toggle Thumbnails Select A Maturity Detection Method for Hemerocallis citrina Baroni Based on Improved YOLOv5 SHENG Bin Journal of Henan Agricultural Sciences    2024, 53 (8): 145-153.   DOI: 10.15933/j.cnki.1004-3268.2024.08.016 Abstract （1872）      PDF （3862KB）（956）       Knowledge map    Save To unify identification standards and improve the detection accuracy and real‑time performance of mature Hemerocallis citrina Baroni picking，an improved GCS‑BI YOLOv5 image detection algorithm was proposed.Firstly，the Ghost lightweight neural networks were utilized to streamline the model structure and save computational resources.Secondly，in order to pay attention to the image channel information and position information simultaneously，efficient attention mechanisms，namely convolutional block attention module（CBAM）and squeeze‑and‑excitation（SE），were cross‑introduced to improve the image feature perception ability and model convergence speed.Then，a weighted bi‑directional feature pyramid network（BI FPN）was used to fuse the multi‑scale image information and improve the comprehensive detection performance of the model for different targets.The experimental results showed that compared with the original algorithm，the lightweight metrics such as the model volume，network layers，number of parameters，and floating‑point operation of the improved algorithm were reduced by 62.89%，33.12%，63.01%，68.39%，respectively.The performance metrics such as detection accuracy and recall rate were improved by 7.77，6.28 percentage points，respectively.Real‑time detection performance was improved by 33.81 f/s.It can be seen that the improved algorithm has better comprehensive performance and can meet the requirements of Hemerocallis citrina Baroni maturity detection. Related Articles | Metrics Select Research on Insect‑bitten Zijin Tea Detection Method Based on YOLOv5s‑SE and Channel Pruning DAI Jiabing, SONG Chunfang, LING Caijin, LI Zhenfeng, SUN Chonggao Journal of Henan Agricultural Sciences    2024, 53 (5): 157-163.   DOI: 10.15933/j.cnki.1004-3268.2024.05.017 Abstract （1112）      PDF （3085KB）（380）       Knowledge map    Save In order to achieve rapid and accurate identification of insect‑bitten Zijin tea leaves in complex nature backgrounds，a detection method for Zijin tea based on YOLOv5s‑SE and channel pruning was proposed. Firstly,SE modules were added to the backbone network of YOLOv5s to enhance the model’s feature extraction capability and reduce interference from complex backgrounds during tea leaf feature extraction.Then,a channel pruning algorithm was used to prune the model and fine‑tuning was conducted,enabling fast and accurate detection of insect‑bitten Zijin tea leaves. Compared to YOLOv5s,the test results showed that the pruned model reduced parameters by 60.1%,improved FPS by 18.6%,reduced GFLOPs by 29.7%,and achieved mAP of 81.3%. Reference | Related Articles | Metrics Select Fine‑Grained Flower Image Classification Based on Neural Network Architecture Search ZHENG Xingkai, YANG Tiejun, HUANG Lin Journal of Henan Agricultural Sciences    2024, 53 (5): 164-171.   DOI: 10.15933/j.cnki.1004-3268.2024.05.018 Abstract （1617）      PDF （3807KB）（349）       Knowledge map    Save To enhance the automation of deep convolutional neural network(CNN)design and improve fine‑grained flower image classification accuracy，an advanced neural network search approach based on differentiable architecture search(DARTS) was proposed.This method automatically constructed fine‑grained flower image classification models.Initially,an attention‑convolution module was constructed to create a comprehensive attention‑convolution search space，thereby increasing the network’s focus on discriminative features. Subsequently,a densely connected reduction cell(DCR cell)with more shallow feature input nodes was developed to retain additional shallow feature information，reducing the loss of discriminative feature information and promoting multi‑scale feature fusion.Lastly,the positions of DCR cells were adjusted when stacking the best cells to create network models of varying parameter sizes,enabling deployment on a broader range of terminal devices.The results showed that this method took approximately 4.5 hours to find the optimal neural network model,achieving classification accuracies of 96.14% on the Oxford 102 dataset and 94.12% on the Flower 17 dataset.Compared with methods like AGNAS,it improved accuracy by 1.40 percentage points on Oxford 102 and 3.09 percentage points on Flower 17. Reference | Related Articles | Metrics Select Light Weight Detection Algorithm for Apple Surface Defect Based on Improved YOLOv7 LI Dahua, KONG Shu, LI Dong, YU Xiao Journal of Henan Agricultural Sciences    2024, 53 (3): 141-150.   DOI: 10.15933/j.cnki.1004-3268.2024.03.015 Abstract （1528）      PDF （4276KB）（1077）       Knowledge map    Save Aiming at how to improve the detection speed and accuracy of apple surface defects and solve the problem of large model memory ratio，a lightweight detection algorithm for apple surface defects based on improved YOLOv7 was proposed. Firstly，GhostNetV2 was introduced as the backbone of YOLOv7 network，which effectively reduced the model complexity and improved the detection speed.SimAM attention‐free mechanism was introduced to enhance the feature information of different depth.The bidirectional weighted feature pyramid BiFPN was used for weighted feature fusion to further improve the detection accuracy of apple surface defects.Finally，the ECIOU loss function was used to calculate the boundary frame loss，which further improved the convergence speed and the overall performance of the model.Experimental results showed that compared with the original YOLOv7 network，the improved model improved the apple surface defect detection mAP@0.5 by 2 percentage points，the accuracy rate and recall rate by 1.7 and 3.9 percentage points respectively. The model decreased by 20.8 MB and the speed increased by 36.43 FPS.Its comprehensive performance was also better than SSD，CenterNet and other mainstream algorithms，which can realize the rapid and accurate diagnosis of apple surface defects. Reference | Related Articles | Metrics Select Lightweight YOLO v5s Blueberry Detection Algorithm Based on Attention Mechanism LIU Yongmin, ZHANG Wei, MA Haizhi, LIU Yuan, ZHANG Yi Journal of Henan Agricultural Sciences    2024, 53 (3): 151-157.   DOI: 10.15933/j.cnki.1004-3268.2024.03.016 Abstract （1915）            Knowledge map    Save To achieve precise and rapid detection of blueberries in natural environments，an improved algorithm combining lightweight networks and attention mechanisms was proposed based on YOLO v5s.Firstly，the structure of the maximum object detection layer was removed at the positions of the backbone network and detection heads，thereby reducing the number of model parameters and enhancing the model’s ability to detect small targets. Secondly，MHSA（Multi‐head self‐attention）was used to replace the C3 module before SPPF（Spatial pyramid pooling‐fast），enabling the model to learn more comprehensive feature representations and enhancing its understanding of complex spatial relationships and contextual information in blueberry images. Finally，S‐PSA（Sequential polarized self‐attention）was added to the C3 module to better capture the contextual dependencies between adjacent regions in the feature map. The experimental results showed that the improved YOLO v5s algorithm improved the detection accuracy of mature blueberries，semi mature blueberries，and immature blueberries by 1.2，4.4，2.6 percentage points，respectively，with average accuracy increase of 2.7 percentage points and 76% reduction in model parameter count. Compared with the current mainstream lightweight object detection models，the improved model has superior performance and can provide an effective solution for the visual system of blueberry picking robots in natural environments. Reference | Related Articles | Metrics Select Identification and Detection of Wheat Kernels with Different Volume Weight Based on Improved U‑Net LÜ Zongwang, WANG Yuqi, SUN Fuyan Journal of Henan Agricultural Sciences    2023, 52 (10): 141-152.   DOI: 10.15933/j.cnki.1004-3268.2023.10.015 Abstract （1419）      PDF （4539KB）（237）       Knowledge map    Save Volume weight is a very important index in the process of wheat quality grade detection.Manual detection and traditional image processing methods have problems such as expensive equipment and low recognition efficiency in wheat quality grade detection，which need to be further improved.The self‑made three grades of wheat grain samples were used as the wheat volume weight dataset，and the U‑Net network was improved according to the characteristics of small grain targets and unclear edge segmentation.On the backbone network，the residual stacking module was used to reduce the feature loss，the CBAM attention mechanism module was embedded in the network bridging part to enhance the further extraction of the features，and the self‑attention mechanism module was embedded in the decoder part to restore the detail information. The results showed that MIoU of the improved network model CBSA_U‑Net was 81.5%，which was 1.8 percentage points higher than U‑Net model，4.2 percentage points higher than PSPNet and 3.3 percentage points higher than DeepLabv3+ model. Reference | Related Articles | Metrics Select Research on Improved U‑Net Method for Rice Leaf Cell Segmentation WEI Gaixing, YI Wenlong, LIU Yucheng, ZHAO Yingding, CHEN Tingzhuo Journal of Henan Agricultural Sciences    2023, 52 (3): 153-160.   DOI: 10.15933/j.cnki.1004-3268.2023.03.017 Abstract （1424）      PDF （3600KB）（470）       Knowledge map    Save Aiming at the issues of blurred boundaries，low signal‑to‑noise ratio，mutual adhesion and stacking of mesophyll cells in the image of rice leaf cells，which lead to low segmentation accuracy，we propose an improved U‑Net method for rice leaf cell segmentation. Firstly，the bridge attention（BA）module is introduced into the ResNeXt network to form the BAResNeXt module as the network encoder to improve the network’s attention to mesophyll cells when extracting deep semantic features；secondly，a channel cross‑attention mechanism is added between the encoder and the decoder to ease the semantic ambiguity between the decoder and the encoder to enhance the information fusion of the segmented image features；finally，the SE attention mechanism is used in the upsampling phase of decoder to filter the interference information of image background. In order to verify the effectiveness of the method，it was compared with deep learning networks such as U‑Net，Res‑UNet，U‑Net++ and Deeplabv3+.The results showed that our method had the best performance in rice leaf cell segmentation. The Precision（96.03%），Recall（97.67%），Acc（97.47%），IoU（93.96%）and Dice（96.78%）of our method were all higher than other networks. Reference | Related Articles | Metrics