Search

Climate change is leading to higher plant water requirements and rootstock can play a role in tree adaptation, since the more vigorous ones are also likely to be more stress resistant. Pear trees of the cv. Abbé Fetél grafted on BA29 (more vigorous) and SYDO (more dwarfing) quince were irrigated according to three different treatments: 110 C, 80 DI and 60 DI, corresponding to 110%, 80% and 60% of the crop evapotranspiration rate (ETc), respectively. Shoot and fruit growth, water potentials, leaf gas exchanges and dry matter content were monitored during the season. Fruit quality was evaluated at harvest and after 6 months of storage at 1 °C. Results show how for both rootstocks, 60 DI significantly decreased their stem (Ψstem) and leaf (Ψleaf) water potentials as well as leaf gas exchanges. In SYDO, final fruit size was affected by irrigation, with lower values on 60 DI, but in BA29, no differences were found between treatments. After storage, BA29 60 DI fruit showed a higher soluble solid content, while in SYDO fruit, firmness was more affected by irrigation level. In conclusion, despite a slight decrease in fruit size, reduced irrigation led to fruit with higher quality features that were also maintained after a long period of storage.

Image processing through the implementation of manually coded algorithms has been adopted to detect fruit damage during post-harvest operations. This study tested convolution neural networks with “You Only Look Once” (YOLO) architecture using a commercial online platform to detect physiological disorders and ripening stage in ‘Abbe Fetel’ pear. Disorders such as superficial scald and the starch pattern index (SPI) were assessed. Three different models were trained to detect: I) individual fruit within the boxes; II) superficial scald or senescence scald on pear skin; III) the SPI value of pears was assessed using the Lugol solution. Preliminary statistics show that the model to count the fruit inside the boxes reached 64.70% of true positives with 0.5 intersection over union. The second had less accuracy (up to 20% of true positives) but maintained a good average precision (60%) with different confidence thresholds (40% and 20%). The third showed good performances compared to the Ctifl and Laimburg scales, with an F1 score of 0.36 and 0.59, respectively. The effectiveness of the transfer learning method was demonstrated. However, further image labelling and modelling research is needed to improve the accuracy of the simulations and to develop an application for portable devices for pre- and post-harvest factor mapping. These results could lead to improvements in the management of fruit boxes and thus help ensure good fruit quality for consumers.

Image processing has recently been adopted for fruit damage detection in post-harvest operations. Through the implementation of hard-coded feature extraction algorithms, high accuracy has been found. The present study tested the fast and operational convolution neural networks with “YOLO v3” architecture using the online platform Supervise.ly to detect on pear fruit ‘Abbe Fetel’ physiological disorders such as superficial scald. Two different models were trained: I) one to detect the individual pear fruits within the batches; II) one to detect superficial scald or senescence scald on pear skin. Preliminary statistics show that the model to count the fruit inside the batches reaches an accuracy of 64.70% with a 0.5 of Intersection of Units. The second one has less accuracy (up to 20% of true positive) but maintains a good level of average precision (0.6) with different confidence thresholds (0.4 and 0.2). Further research is needed to improve the accuracy of both models and to map quality pre- and post-harvest. These results will help the packing house to manage fruit batches and to ensure good fruit quality for consumers.

Image/video processing for fruit detection in the tree using hard-coded feature extraction algorithms has shown high accuracy on fruit detection during recent years. While accurate, these approaches even with high-end hardware are still computationally intensive and too slow for real-time systems. This paper details the use of deep convolution neural networks architecture based on single-stage detectors. Using deep-learning techniques eliminates the need for hard-code specific features for specific fruit shapes, color and/or other attributes. This architecture takes the input image and divides into AxA grid, where A is a configurable hyper-parameter that defines the fineness of the grid. To each grid cell an image detection and localization algorithm is applied. Each of those cells is responsible to predict bounding boxes and confidence score for fruit (apple and pear in the case of this study) detected in that cell. We want this confidence score to be high if a fruit exists in a cell, otherwise to be zero, if no fruit is in the cell. More than 100 images of apple and pear trees were taken. Each tree image with approximately 50 fruits, that at the end resulted on more than 5000 images of apple and pear fruits each. Labeling images for training consisted on manually specifying the bounding boxes for fruits, where (x, y) are the center coordinates of the box and (w, h) are width and height. This architecture showed an accuracy of more than 90% fruit detection. Based on correlation between number of visible fruits, detected fruits on one frame and the real number of fruits on one tree, a model was created to accommodate this error rate. Processing speed is higher than 20 FPS which is fast enough for any grasping/harvesting robotic arm or other real-time applications. HIGHLIGHTS: Using new convolutional deep learning techniques based on single-shot detectors to detect and count fruits (apple and pear) within the tree canopy.