The detection of Lingwu long jujubes in a natural environment is of great significance for robotic picking. Therefore, a lightweight network of target detection based on the SSD (single shot multi-box detector) is presented to meet the requirements of a low computational complexity and enhanced precision. Traditional object detection methods need to load pre-trained weights, cannot change the network structure, and are limited by equipment resource conditions. This study proposes a lightweight SSD object detection method that can achieve a high detection accuracy without loading pre-trained weights and replace the Peleenet network with VGG16 as the trunk, which can acquire additional inputs from all of the previous layers and provide itself characteristic maps to all of the following layers. The coordinate attention module and global attention mechanism are added in the dense block, which boost models to more accurately locate and identify objects of interest. The Inceptionv2 module has been replaced in the first three additional layers of the SSD structure, so the multi-scale structure can enhance the capacity of the model to retrieve the characteristic messages. The output of each additional level is appended to the export of the sub-level through convolution and pooling operations in order to realize the integration of the image feature messages between the various levels. A dataset containing images of the Lingwu long jujubes was generated and augmented using pre-processing techniques such as noise reinforcement, light variation, and image spinning. To compare the performance of the modified SSD model to the original model, a number of experiments were conducted. The results indicate that the mAP (mean average precision) of the modified SSD algorithm for object inspection is 97.32%, the speed of detection is 41.15 fps, and the parameters are compressed to 30.37% of the original networks for the same Lingwu long jujubes datasets without loading pre-trained weights. The improved SSD target detection algorithm realizes a reduction in complexity, which is available for the lightweight adoption to a mobile platform and it provides references for the visual detection of robotic picking.

In the mechanized harvesting of Lycium barbarum L. (L. barbarum), there are prominent problems such as low harvesting efficiency, high damage rate, incomplete separation of leaves and delayed transportation. Therefore, an integrated L. barbarum harvester was designed and developed in this study, which has the functions of picking, undertaking, transportation, winnowing and collection. The design requirements and constraints were identified by cultivation agronomy. Through simulation and physical tests, the tarpaulin was determined as the undertaking material. This machine achieved efficient picking with a vibrating picker with a multi-degree-of-freedom picking arm. The two-stage conveyor belts and the intermediate receiving plate were designed for low loss transportation of fruit. The axial flow fan and secondary buffer device were used to realize winnowing and reduce the damage rate. Through the three-factor and three-level orthogonal test, an optimal working parameter combination was determined: the vibration frequency of the picker was 20 Hz, the conveyor speed was 4 m/min, the airflow speed of the fan was 7 m/s. A field test was conducted under these parameters, and the results showed that the harvesting efficiency was about five times that of manual harvesting. The integrated L. barbarum harvester basically met the harvesting requirements and provided a new scheme for mechanized harvesting.

In order to reduce the impact of pests and diseases on the yield and quality of Ningxia wine grapes and to improve the efficiency and intelligence of detection, this paper designs an intelligent detection platform for pests and diseases. The optimal underlying network is selected by comparing the recognition accuracy of both MobileNet V2 and YOLOX_s networks trained on the Public Dataset. Based on this network, the effect of adding attention mechanism and replacing loss function on recognition effect is investigated by permutation in the Custom Dataset, resulting in the improved network YOLOX_s + CBAM. The improved network was trained on the Overall Dataset, and finally a recognition model capable of identifying nine types of pests was obtained, with a recognition accuracy of 93.35% in the validation set, an improvement of 1.35% over the original network. The recognition model is deployed on the Web side and Raspberry Pi to achieve independent detection functions; the channel between the two platforms is built through Ngrok, and remote interconnection is achieved through VNC desktop. Users can choose to upload local images on the Web side for detection, handheld Raspberry Pi for field detection, or Raspberry Pi and Web interconnection for remote detection.