Accurate and rapid crack detection has been one of the most important steps in duck eggs before processing. The shell of duck eggs may crack, particularly when subjected to the vibration, extrusion, and collision during production, transportation, packaging and storage. The microorganism can then invade through the crack, easy to cause the deterioration of duck eggs. Furthermore, the damage of one duck egg can greatly affect the processing requirements of the whole batch. However, manual detection cannot fully meet the large-scale production in recent years, due to the unstable accuracy rate, low efficiency and industrialization in the poultry egg industry at present. In this study, a lightweight detection was proposed for the multi-column cracks of duck eggs using improved YOLOv5l (you only look once version5 large). The better detection of duck egg was achieved in the simple, small amount of calculation and storage. A LED light was used to illuminate the duck egg under dark conditions. There was the difference of light transmittance between cracked eggshell and intact eggshell. Specifically, the Ghost_conv module was introduced to replace the original convolution in YOLOv5. The calculation and parameters of model were greatly reduced to increase the training speed with more lightweight. In addition, ECA (efficient channel attention) attention mechanism was added to the backbone Network and BIFPN (bi-directional feature pyramid network). The effective information was detected using the attention mechanism. Among them, the BIFPN was used to extract and fuse some features at different scales. The accuracy of model was improved to reduce the missing and wrong detection. At the same time, the fusion of CIoU and α-IoU loss function was used to replace the original GIoU function of YOLOv5, thus accelerating the regression prediction. A dataset of duck egg was obtained with 892 images, where a camera was used to capture duck eggs rolling on a conveyor. The dataset was then enhanced by random cropping, adding noise and mixup, in order to further improve the generalization and robustness of the model. The self-developed dataset of duck egg cracks was utilized to verify the detection performance of the improved model. The results show that the improved YOLOv5l network model shared the high performance. The detection accuracy of the improved model on the test set was 93.8%, which was 6.3 percentage points higher than that of the original YOLOv5l model. The number of parameters, floating point arithmetic and the occupied storage was reduced by 30.6%, 39.4% and 34.6%, respectively. The frame rate of detection was 28.954 frames /s, which was only 3.824 frames /s lower than the original YOLOv5l model. The detection accuracy of the improved YOLOv5l increased by 13.1, 12.5 and 8.2 percentage points, respectively, compared with the common target detection networks, such as single shot MultiBox detector (SSD), YOLOv4, and faster-RCNN (faster region convolutional neural networks). The high-precision detection with the low hardware resources can be expected to provide the promising solutions and technical applications.

Melanin grading has been widely used in the epidermal layer of live yellow feather broilers. However, it is in high demand to improve the efficiency, cost-saving and susceptibility to lighting conditions. This study aims to explore an intelligent melanin grading (ConvNeXt-WPCA) for the epidermal layer of live yellow feather broilers using the ConvNeXt model. Three key enhancements were proposed to improve the ConvNeXt-WPCA model for the recognition of melanin in broilers. Firstly, the channel weights of the input images were adjusted to treat the uneven distribution of melanin across the RGB channels in the melanin images of yellow feather broilers. The channel with more melanin was then emphasized to extract the melanin features. The reweighting of channels was used to more effectively aggregate the melanin signals for the better classification performance of the deep learning model. Secondly, Depthwise Separable Convolution (DWConv) was replaced with the partial convolution. The computational load and memory access times were reduced to improve the utilization of computational resources. Lastly, the Coordinate Attention (CA) module was introduced to focus on the key skin regions near the chest and anus of yellow feather broiler, thereby improving the classification accuracy of models. At the same time, a dual-light source image acquisition device was designed to efficiently and simultaneously collect images under both normal and polarized lighting conditions. Sufficient data was available for model training and performance evaluation. There was a minimum impact of lighting conditions on grading. Furthermore, the potential application of polarized light was also explored in the tasks of melanin grading. The results demonstrated that the ConvNeXt-WPCA model was improved by 9.68 percentage points in the grading accuracy rate of the melanin image dataset for the yellow feather broiler under natural light, compared with the standard ConvNeXt model. A final recognition accuracy rate reached 89.03%. The grading accuracy rate was improved by 15.26 percentage points in the polarized light ones, with a final recognition accuracy rate of 98.87%. Moreover, both the parameters and floating-point volume were reduced. In conclusion, the melanin image grading of the yellow feather broiler epidermal layer under polarized light conditions was superior to that under natural light. The high recognition accuracy was achieved in the ConvNeXt-WPCA melanin grading for the epidermal layer of yellow feather broilers. This finding can provide a theoretical basis and technical support for the practical application of intelligent melanin grading in the epidermal layer of yellow feather broilers. Significant implications were obtained for the accurate melanin grading of broilers in poultry industry. The ConvNeXt-WPCA model improved the accuracy of melanin grading with the required computational resources, indicating the practical and efficient solution to real-world applications. The model can also be further optimized for the application in other poultry species.

Object detection of small intestinal villus in pathology images of pigs is of practical use in the agricultural and veterinary fields. However, manual detection cannot fully meet the large-scale production at present, due to the time-consuming, and labor-intensive task. In this study, an improved YOLOv5s model was proposed to rapidly and accurately detect the specific location of small intestinal villus in pathology images. The source of experimental data was taken as the pathology images of small intestinal villus in pigs provided by the Institute of Animal Husbandry, Guangdong Academy of Agricultural Sciences. To train the model, 64 high-resolution images of intact small intestinal villus were also collected and then cropped into 626 images with a size of 640×640. According to the ratio of 8:2, these images were divided into the training and validation sets. Firstly, the spatial pyramid was optimized in the backbone network of YOLOv5s using hybrid pooling in the form of tandem. The feature extraction and representation further improved the detection accuracy. This optimization was used to better capture the feature information at different scales. The model was trained to more accurately learn the content. Secondly, a network module called SimAM (A Simple, Parameter-Free Attention Module) was introduced to combine the attention mechanism and residual connectivity of the Bottleneck. A weighted feature representation was then obtained to weight the feature maps in Bottleneck. Thereby the model was enhanced to perceive the target. Compared with the traditional attention module, the SimAM module featured simplicity without the parameter tuning, leading to a more efficient model and easy to implement. The spatial pyramid and the attention mechanism after improvement were better utilized to enhance the detection performance of small intestinal villus in pigs. This improved approach reduced the model parameters and computation with the high detection accuracy, indicating a more lightweight and efficient model. The enhancing ability to perceive the target was used to more accurately locate and identify small intestinal villus in pathology images. More reliable tools and support were provided in the agricultural and veterinary fields. According to the experimental results, the improved model achieved significant performance in detecting the small intestinal villus in pathology images of pigs. The average precision value reached 92.43%, while the detection speed was 40 frames per second. Compared with the original YOLOv5s model, the improved model reduced the inference time by 1.04 ms per frame without increasing the number of parameters. Also, the average precision was improved by 4.62 percentage points. Compared with other common target detection models, significant advantages were found in the detection accuracy, detection speed, and the number of model parameters. There were significant improvements in both speed and accuracy. The network structure was optimized to introduce an attention mechanism, in order to more accurately detect the small bowel villus in pathology images with the higher detection speed. In addition, the improved model was also more efficient and feasible for practical applications, in terms of the number of parameters.

A near-real-time recognition model was constructed to rapidly and timely identify the rice growth stages using sequential spectral images. The key stages of rice growth were identified, including tillering, jointing, heading, milking, and maturity. An optimal combination of features was also obtained. The rice experimental field was carried out in the Tianhe District, Guangzhou City, Guangdong Province, China. Unmanned aerial vehicles (UAVs) were involved in capturing spectral images of the field. The drone imaging system was equipped with one visible light and five multispectral cameras, each of which had a 2-million-pixel resolution for the high-definition captures. Detailed spectral data was obtained to precisely identify the different rice growth stages. A significant challenge was the variation in the RGB values under different lighting conditions. The reason was that the accurate segmentation of the rice canopy was hindered by using color as the primary feature. A novel spectral image segmentation was introduced to combine the Relative Vegetation Index using an iterative self-organizing data analysis. The complete area of the rice canopy was precisely extracted for the accurate identification of the growth stage. Then, the dataset was simplified to reduce the dimensions and computational complexity, in order to enhance the performance and efficiency of recognition. Among them, the expression of features was analyzed during different rice growth stages, where the reflectance data was from green, red, blue, red-edge, and near-infrared bands. Eight vegetation indices were considered as the candidate features, including the normalized difference red edge index (NDRE), visual atmospherically resistant index (VARI), and red edge chlorophyll index (RECI). Minimum Redundancy Feature Selection was used to efficiently rank and select the most relevant and least redundant features, in order to avoid computational overload and potential impact on the model. The optimal combination of features was identified as the NDRE, VARI, and RECI, thus forming the crux of the feature engineering. Finally, this combination was integrated into the Extreme Gradient Boosting to construct an advanced recognition model of the rice growth stage. The improved model with the Extreme Gradient Boosting also exhibited exceptional performance with minimal confusion among the five growth stages. The accuracy rates of the model were 98.08%, 100.00%, 99.68%, 97.50%, and 99.29%, respectively, for the tillering, jointing, heading, milking, and maturity stages of rice. An overall recognition accuracy of 98.77% was achieved. Additionally, the improved model scored a Kappa coefficient of 0.984 and a macro-F1 score of 0.989 1. The XGBoost improved the overall recognition accuracy, Kappa coefficient, and F1 score by 1.59 percentage points, 0.020, and 0.015, respectively, compared with the SVM. While the improvements were 1.23 percentage points, 0.015, and 0.011, respectively, compared with RF. The finding can provide a reliable, near-real-time, and highly accurate identification of rice growth stages. This improved model can also offer essential support for decision-making in agricultural practices, particularly in crop production and field management for the high crop yield in precision agriculture.

Spike development directly affects the yield and quality of rice. We describe an algorithm for automatically identifying multiple developmental stages of rice spikes (AI-MDSRS) that transforms the automatic identification of multiple developmental stages of rice spikes into the detection of rice spikes of diverse maturity levels. The scales vary greatly in different growth and development stages because rice spikes are dense and small, posing challenges for their effective and accurate detection. We describe a rice spike detection model based on an improved faster regions with convolutional neural network (Faster R-CNN). The model incorporates the following optimization strategies: first, Inception_ResNet-v2 replaces VGG16 as a feature extraction network; second, a feature pyramid network (FPN) replaces single-scale feature maps to fuse with region proposal network (RPN); third, region of interest (RoI) alignment replaces RoI pooling, and distance-intersection over union (DIoU) is used as a standard for non-maximum suppression (NMS). The performance of the proposed model was compared with that of the original Faster R-CNN and YOLOv4 models. The mean average precision (mAP) of the rice spike detection model was 92.47%, a substantial improvement on the original Faster R-CNN model (with 40.96% mAP) and 3.4% higher than that of the YOLOv4 model, experimentally indicating that the model is more accurate and reliable. The identification results of the model for the heading–flowering, milky maturity, and full maturity stages were within two days of the results of manual observation, fully meeting the needs of agricultural activities.