Unmanned vehicle for spraying is used in production of facility agriculture and smart agriculture. Tomato in greenhouse has characteristics of narrow planting spacing and many twisted lead wires, which required special vehicles. Therefore, it is necessary to design a miniaturized, variable, and intelligent spraying unmanned vehicle for greenhouse tomato. It is mainly composed of a disease detection and positioning module, a lifting platform, a steering gear rocker mechanism, and a fully automatic bearing chassis. This design innovatively combines deep learning technology to achieve automatic detection of disease targets. RGB image of greenhouse tomato collected by Kinect V2 was used as input for disease detection. Two-dimensional pixel coordinates of the disease detection results were converted into three-dimensional spatial coordinates for realizing disease location. Ball screw and stepping motor were used to adjust height of the lifting platform. Then, the steering gear rocker mechanism was employed to locate disease of the greenhouse tomato. Variable and intelligent spraying was thus completed. Results showed that target detection accuracy of fruit clusters and diseases in complex environment reached 75.15% with F1 score of 79.96 through testing the trained YOLOv4-tiny model. It detected 37.6 images per second, so it was able to be deployed on the embedded development board to detect disease targets of greenhouse tomato. Compared with the results of manual disease location, results showed that absolute error of disease location was within ±3.5 cm. Through field test of the spraying unmanned vehicle, the success rate of the whole machine of the intelligent spraying unmanned vehicle for greenhouse tomato was above 75%, and the error rate of spraying pesticide was below 20%, which achieved accurate positioning of diseases of greenhouse tomato and variable intelligent spraying according to disease degree. This design can provide reference for the design of other spraying unmanned vehicle and has a good promotion and application prospect.

The coefficient of restitution (CoR) is an important parameter for designing vibration-harvesting machinery. There are three main types of fruit-to-fruit collisions during vibration harvesting: collision between fruits collected using a collection device and falling fruits, collision between fruits on branches before being removed, and collision of fruits in the air. The CoR for the first two types of collision was investigated separately using drop and pendulum methods. However, there have been few studies on CoR for the collision of fruits in the air. In this study, a platform was designed to simulate the collision of fruits in the air during vibration harvesting for the ‘Gala’ apple, where influences of collision velocity on CoR were studied. Images from a high-speed camera were processed based on RGB to Lab conversion to extract the bruise surface and calculate the bruise volume. Total bruise volume, the sum of two apple bruise volumes, was calculated and analyzed in relation to the CoR. Results showed that the CoR decreased with collision velocity increasing from 1.0 m/s to 1.4 m/s, where the CoR reached 0.93 or higher when collision velocity was 1.0 m/s, making fruits not bruise, while fruits began to bruise when collision velocity increased from 1.2 m/s. The CoR did not continue to decrease when collision velocity exceeded 1.4 m/s due to rotation. There was little correlation between total bruise volume and the CoR due to the composite motion of fruits in the air, indicating that the CoR may not be an indicator to determine the degree of fruit bruise when the fruit made a composite motion during the collision. Therefore, this research is expected to guide the establishment of a more accurate fruit model to design optimal vibration harvesting machinery.

Recent approaches to the internal quality inspection of apples with the application of hyperspectral imaging technology are highly cost-intensive because of labor involvement for the data collection on a fixed posture and manual selection of the region of interest (RoI). In addition, several studies have repeated the data acquisition for the same apple. Current methods cannot meet the automation requirements of the sorting line. Therefore, this study proposed a novel method for automatically selecting RoI in hyperspectral images of apples with random poses. Firstly, the preliminary RoI selection of apple hyperspectral image was carried out, followed by the performance of histogram statistics of each pixel with spectral intensity at 700 nm wavelength. The top 40% area of the spectral intensity was reserved to obtain the magnitude relationship of the spectral intensity of each pixel point and a morphological erosion operation. Original apple RoI was acquired and overexposed pixels were removed with spectral intensity greater than 3900 (maximum 4095) in the reserved area at 700 nm. Secondly, the relationship between apple size and prediction accuracy was measured for the in-depth RoI analysis. A partial least square regression (PLSR) model was established between the average spectrum and apple sugar content of RoI with different sizes. Finally, the established model with the top 70% of the spectral intensity achieved the best prediction accuracy. Non-destructive estimation of apple sugar content was performed through hyperspectral imaging technology with reference to the proposed RoI selection method. A competitive adaptive reweighted sampling algorithm along the PLSR (CARS-PLSR) model was established after black-and-white correction and standard normal transformation (SNV) preprocessing and obtained the highest prediction accuracy. The determination coefficient of cross-validation (R_cv) and root mean square error of cross-validation (RMSECV) were 0.9595 and 0.3203°Brix, respectively. The determination coefficient of prediction (R_p) was 0.9308, and the root mean square error of prediction (RMSEP) was 0.4681°Brix. Results proved that the auto-selection of RoI is an efficient and accurate method, which can provide a foundation in practical application for online apple grading systems with hyperspectral imaging technology.