A pepper quality detection and classification model based on transfer learning combined with convolutional neural network is proposed as a solution for low efficiency of manual pepper sorting at the current stage. The pepper dataset was amplified with data pre-processing methods including rotation, luminance switch, and contrast ratio switch. To improve training speed and precision, a network model was optimized with a fine-tuned VGG 16 model in this research, transfer learning was applied after parameter optimization, and comparative analysis was performed by combining ResNet50, MobileNet V2, and GoogLeNet models. It turned out that the VGG 16 model output anticipation precision was 98.14%, and the prediction loss rate was 0.0669 when the dropout was settled as 0.3, learning rate settled as 0.000001, batch normalization added, and ReLU as activation function. Comparing with other finetune models and network models, this model was of better anticipation performance, as well as faster and more stable convergence rate, which embodied the best performance. Considering the basis of transfer learning and integration with strong generalization and fitting capacity of the VGG 16 finetune model, it is feasible to apply this model to the external quality classification of pepper, thus offering technical reference for further realizing the automatic classification of pepper quality.

To detect quickly and accurately “Yuluxiang” pear fruits in non-structural environments, a lightweight YOLO-GEW detection model is proposed to address issues such as similar fruit color to leaves, fruit bagging, and complex environments. This model improves upon YOLOv8s by using GhostNet as its backbone for extracting features of the “Yuluxiang” pears. Additionally, an EMA attention mechanism was added before fusing each feature in the neck section to make the model focus more on the target information of “Yuluxiang” pear fruits, thereby improving target recognition ability and localization accuracy. Furthermore, the CIoU Loss was replaced with the WIoUv3 Loss as the loss function, which enhances the capability of bounding box fitting and improves model performance without increasing its size. Experimental results demonstrated that the enhanced YOLO-GEW achieves an F1 score of 84.47% and an AP of 88.83%, while only occupying 65.50% of the size of YOLOv8s. Compared to lightweight algorithms such as YOLOv8s, YOLOv7-Tiny, YOLOv6s, YOLOv5s, YOLOv4-Tiny, and YOLOv3-Tiny; there are improvements in AP by 2.32%, 1.51%, 2.95%, 2.06%, 2.92%, and 5.38% respectively. This improved model can efficiently detect “Yuluxiang” pears in non-structural environments in real-time and provides a theoretical basis for recognition systems used by picking robots.

The intelligent detection of young peaches is the main technology of fruit-thinning robots, which is crucial for enhancing peach fruit quality and reducing labor costs. This study presents the lightweight YOLO-PEM model based on YOLOv8s to achieve high-precision and automatic detection of young “Okubo” peaches. Firstly, the C2f_P module was devised by partial convolution (PConv), replacing all C2f modules in YOLOv8s to achieve the model’s lightweight. Secondly, embedding the efficient multi-scale attention (EMA) module in the lightweight C2f_P_1 module of the backbone network enhanced the feature extraction capability and accuracy for young peaches. Finally, the MPDIoU loss function was utilized to replace the original CIoU loss function, which improved the detection accuracy of the bounding box while speeding up the convergence of the model. The experimental results demonstrate that the YOLO-PEM model achieved an average precision (AP) of 90.86%, F1 score of 86.70%, and model size of 16.1 MB, which was a 1.85% improvement in the AP, 0.85% improvement in the F1 score, and 5.3 MB reduction in the model size compared with YOLOv8s. The AP was 6.26%, 6.01%, 2.05%, 2.12%, and 1.87% higher compared with the other lightweight detection models YOLOv3-tiny, YOLOv4-tiny, YOLOv5s, YOLOv6s, and YOLOv7-tiny, respectively. Furthermore, the FPS of YOLO-PEM was 196.2 f·s^-1, which can fulfill the demand for the real-time detection of young peaches. YOLO-PEM effectively detects young peaches in complex orchard environments and can offer a basis for the theoretical design of the vision system of the “Okubo” peach fruit-thinning robot and scientific management of orchards.

The efficient identification of the field flat jujube is the first condition to realize its automated picking. Consequently, a lightweight algorithm of target identification based on improved YOLOv5 (you only look once) is proposed to meet the requirements of high-accuracy and low-complexity. At first, the proposed method solves the imbalance of data distribution by improving the methods of data enhancement. Then, to improve the accuracy of the model, we adjust the structure and the number of the Concentrated-Comprehensive Convolution Block modules in the backbone network, and introduce the attention mechanisms of Efficient Channel Attention and Coordinate Attention. On this basis, this paper makes lightweight operations by using the Deep Separable Convolution to reduce the complexity of the model. Ultimately, the Complete Intersection over Union loss function and the non-maximum suppression of Distance Intersection over Union are used to optimize the loss function and the post-processing process, respectively. The experimental results show that the mean average precision of improved network reaches 97.4%, which increases by 1.7% compared with the original YOLOv5s network; and, the parameters, floating point of operations, and model size are compressed to 35.39%, 51.27%, and 37.5% of the original network, respectively. The comparison experiments are conducted around the proposed method and the common You Only Look Once target detection algorithms. The experimental results show that the mean average precision of the proposed method is 97.4%, which is higher than the 90.7%, 91.7%, and 88.4% of the YOLOv3, YOLOv4, and YOLOx-s algorithms, and the model size decreased to 2.3%, 2.2%, and 15.7%, respectively. The improved algorithm realizes a reduction of complexity and an increase in accuracy, it can be suitable for lightweight deployment to a mobile terminal at a later stage, and it provides a certain reference for the visual detection of picking robots.

Fruits with various maturity levels coexist among the harvested jujubes, and have different tastes and uses. Manual grading has a low efficiency and a strong subjectivity. The number of “Hupingzao” jujubes between different maturity levels is unbalanced, which affects the performance of the classifier. To solve the above issue, the class balance loss (CB) was used to improve the MobileNet V2 network, and a transfer learning strategy was used to train the model. The model was optimized based on the selection of an optimizer and learning rate. The model achieved the best classification results using the AdamW optimizer and a learning rate of 0.0001. The application of transfer learning and class balance loss improved the model’s performance. The precision was 96.800~100.000%, the recall was 95.833~100.000%, and the F1 score was 0.963~1.000. To compare the CB-MobileNet V2 performance, the CB-AlexNet, CB-GoogLeNet, CB-ShuffleNet, CB-Inception V3, CB-ResNet 50, and CB-VGG 16 with transfer learning were used to build classification models. Achieving a validation accuracy of 99.058%, and a validation loss value of 0.055, the CB-MobileNet V2 model showed a better overall performance compared with other models. The maturity detection system of “Hupingzao” jujubes was developed to test the model. The testing accuracy of the CB-MobileNet V2 model was 99.294%. The research indicates that the CB-MobileNet V2 model improves the performance of maturity classification, and provides a theoretical basis for intelligent classification of the quality of “Hupingzao” jujubes.

To solve the failure problem of the visible/near infrared (VIS/NIR) spectroscopy model, soluble solids content (SSC) detection for fresh jujubes cultivated in different modes was carried out based on the method of variable optimization and model update. Iteratively retained informative variables (IRIV) and successive projections algorithm (SPA) algorithms were used to extract characteristic wavelengths, and least square support vector machine (LS-SVM) was used to establish detection models. Compared with IRIV, IRIV-SPA achieved better performance. Combined with the offset properties of the wavelength, repeated wavelengths were removed, and wavelength recombination was carried out to create a new combination of variables. Using these fused wavelengths, the model was recalibrated based on the Euclidean distance between samples. The LS-SVM detection model of SSC was established using the update method of wavelength fusion-Euclidean distance. Good prediction results were achieved using the proposed model. The determination coefficient (R²), root mean square error (RMSE), and residual predictive deviation (RPD) of the test set on SSC of fresh jujubes cultivated in the open field were 0.82, 1.49%, and 2.18, respectively. The R², RMSE, and RPD of the test set on SSC of fresh jujubes cultivated in the rain shelter were 0.81, 1.44%, and 2.17, respectively. This study realized the SSC detection of fresh jujubes with different cultivation and provided a method for the establishment of a robust VIS/NIR detection model for fruit quality, effectively addressing the industry need for identifying jujubes grown in the open field.