RESUMEN
This study addresses the challenges of low detection precision and limited generalization across various ripeness levels and varieties for large non-green-ripe citrus fruits in complex scenarios. We present a high-precision and lightweight model, YOLOC-tiny, built upon YOLOv7, which utilizes EfficientNet-B0 as the feature extraction backbone network. To augment sensing capabilities and improve detection accuracy, we embed a spatial and channel composite attention mechanism, the convolutional block attention module (CBAM), into the head's efficient aggregation network. Additionally, we introduce an adaptive and complete intersection over union regression loss function, designed by integrating the phenotypic features of large non-green-ripe citrus, to mitigate the impact of data noise and efficiently calculate detection loss. Finally, a layer-based adaptive magnitude pruning strategy is employed to further eliminate redundant connections and parameters in the model. Targeting three types of citrus widely planted in Sichuan Province-navel orange, Ehime Jelly orange, and Harumi tangerine-YOLOC-tiny achieves an impressive mean average precision (mAP) of 83.0%, surpassing most other state-of-the-art (SOTA) detectors in the same class. Compared with YOLOv7 and YOLOv8x, its mAP improved by 1.7% and 1.9%, respectively, with a parameter count of only 4.2M. In picking robot deployment applications, YOLOC-tiny attains an accuracy of 92.8% at a rate of 59 frames per second. This study provides a theoretical foundation and technical reference for upgrading and optimizing low-computing-power ground-based robots, such as those used for fruit picking and orchard inspection.
RESUMEN
Introduction: Insect pests from the family Papilionidae (IPPs) are a seasonal threat to citrus orchards, causing damage to young leaves, affecting canopy formation and fruiting. Existing pest detection models used by orchard plant protection equipment lack a balance between inference speed and accuracy. Methods: To address this issue, we propose an adaptive spatial feature fusion and lightweight detection model for IPPs, called ASFL-YOLOX. Our model includes several optimizations, such as the use of the Tanh-Softplus activation function, integration of the efficient channel attention mechanism, adoption of the adaptive spatial feature fusion module, and implementation of the soft Dlou non-maximum suppression algorithm. We also propose a structured pruning curation technique to eliminate unnecessary connections and network parameters. Results: Experimental results demonstrate that ASFL-YOLOX outperforms previous models in terms of inference speed and accuracy. Our model shows an increase in inference speed by 29 FPS compared to YOLOv7-x, a higher mAP of approximately 10% than YOLOv7-tiny, and a faster inference frame rate on embedded platforms compared to SSD300 and Faster R-CNN. We compressed the model parameters of ASFL-YOLOX by 88.97%, reducing the number of floating point operations per second from 141.90G to 30.87G while achieving an mAP higher than 95%. Discussion: Our model can accurately and quickly detect fruit tree pest stress in unstructured orchards and is suitable for transplantation to embedded systems. This can provide technical support for pest identification and localization systems for orchard plant protection equipment.
RESUMEN
Chinese liquor (Baijiu) is one of the four major distilled spirits in the world. At present, liquor products containing impurities still exist on the market, which not only damage corporate image but also endanger consumer health. Due to the production process and packaging technologies, impurities usually appear in products of Baijiu before entering the market, such as glass debris, mosquitoes, aluminium scraps, hair, and fibres. In this paper, a novel method for detecting impurities in bottled Baijiu is proposed. Firstly, the region of interest (ROI) is cropped by analysing the histogram projection of the original image to eliminate redundant information. Secondly, to adjust the number of distributions in the Gaussian mixture model (GMM) dynamically, multiple unmatched distributions are removed and distributions with similar means are merged in the process of modelling the GMM background. Then, to adaptively change the learning rates of the front and background pixels, the learning rate of the pixel model is created by combining the frame difference results of the sequence images. Finally, a histogram of oriented gradient (HOG) features of the moving targets is extracted, and the Support Vector Machine (SVM) model is chosen to exclude bubble interference. The experimental results show that this impurity detection method for bottled Baijiu controls the missed rate by within 1% and the false detection rate by around 3% of impurities. Its speed is five times faster than manual inspection and its repeatability index is good, indicating that the overall performance of the proposed method is better than manual inspection with a lamp. This method is not only efficient and fast, but also provides practical, theoretical, and technical support for impurity detection of bottled Baijiu that has broad application prospects.
RESUMEN
The soluble solid content (SSC) is one of the important parameters depicting the quality, maturity and taste of fruits. This study explored hyperspectral imaging (HSI) and fluorescence spectral imaging (FSI) techniques, as well as suitable chemometric techniques to predict the SSC in kiwifruit. 90 kiwifruit samples were divided into 70 calibration sets and 20 prediction sets. The hyperspectral images of samples in the spectral range of 387 nm~1034 nm and the fluorescence spectral images in the spectral range of 400 nm~1000 nm were collected, and their regions of interest were extracted. Six spectral pre-processing techniques were used to pre-process the two spectral data, and the best pre-processing method was selected after comparing it with the predicted results. Then, five primary and three secondary feature extraction algorithms were used to extract feature variables from the pre-processed spectral data. Subsequently, three regression prediction models, i.e., the extreme learning machines (ELM), the partial least squares regression (PLSR) and the particle swarm optimization - least square support vector machine (PSO-LSSVM), were established. The prediction results were analyzed and compared further. MASS-Boss-ELM, based on fluorescence spectral imaging technique, exhibited the best prediction performance for the kiwifruit SSC, with the R p 2 , R c 2 and RPD of 0.8894, 0.9429 and 2.88, respectively. MASS-Boss-PLSR based on the hyperspectral imaging technique showed a slightly lower prediction performance, with the R p 2 , R c 2 , and RPD of 0.8717, 0.8747, and 2.89, respectively. The outcome presents that the two spectral imaging techniques are suitable for the non-destructive prediction of fruit quality. Among them, the FSI technology illustrates better prediction, providing technical support for the non-destructive detection of intrinsic fruit quality.