A deep neural network approach towards real-time on-branch fruit recognition for precision horticulture

Saedi, Seyed Iman; Khosravi, Hossein

doi:10.1016/j.eswa.2020.113594

Cited by 58 publications

(24 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the fruit industry, visual inspection and image processing for the recognition and positioning of fruits and flowers are among the most intensively studied topics ( Gongal et al, 2015 ; Stein et al, 2016 ; Tang et al, 2020 ). Visual features are used to differentiate between the targets and other objects ( Saedi and Khosravi, 2020 ). Classical image processing algorithms include those based on color, threshold segmentation, and edge detection.…”

Section: Introductionmentioning

confidence: 99%

Multi-Target Recognition of Bananas and Automatic Positioning for the Inflorescence Axis Cutting Point

Duan

Chen

et al. 2021

Front. Plant Sci.

View full text Add to dashboard Cite

Multi-target recognition and positioning using robots in orchards is a challenging task in modern precision agriculture owing to the presence of complex noise disturbance, including wind disturbance, changing illumination, and branch and leaf shading. To obtain the target information for a bud-cutting robotic operation, we employed a modified deep learning algorithm for the fast and precise recognition of banana fruits, inflorescence axes, and flower buds. Thus, the cutting point on the inflorescence axis was identified using an edge detection algorithm and geometric calculation. We proposed a modified YOLOv3 model based on clustering optimization and clarified the influence of front-lighting and backlighting on the model. Image segmentation and denoising were performed to obtain the edge images of the flower buds and inflorescence axes. The spatial geometry model was constructed on this basis. The center of symmetry and centroid were calculated for the edges of the flower buds. The equation for the position of the inflorescence axis was established, and the cutting point was determined. Experimental results showed that the modified YOLOv3 model based on clustering optimization showed excellent performance with good balance between speed and precision both under front-lighting and backlighting conditions. The total pixel positioning error between the calculated and manually determined optimal cutting point in the flower bud was 4 and 5 pixels under the front-lighting and backlighting conditions, respectively. The percentage of images that met the positioning requirements was 93 and 90%, respectively. The results indicate that the new method can satisfy the real-time operating requirements for the banana bud-cutting robot.

show abstract

Section: Introductionmentioning

confidence: 99%

Multi-Target Recognition of Bananas and Automatic Positioning for the Inflorescence Axis Cutting Point

Duan

Chen

et al. 2021

Front. Plant Sci.

View full text Add to dashboard Cite

show abstract

“…The detection speed is faster, and it is more suitable for visual recognition tasks in agricultural scenes. Saedi and Khosravi (2020) aimed at the identification of fruits in unstructured orchards, and proposed a convolutional neural network based on RGB images to realize the detection of six types of fruits: green apple, nectarine, apricot, peach, sour cherry, and amber plum. The model had three convolutional layers.…”

Section: Introductionmentioning

confidence: 99%

An improved Yolov3 based on dual path network for cherry tomatoes detection

Chen

Wang

et al. 2021

J Food Process Engineering

View full text Add to dashboard Cite

With the development of deep learning theory, the application of Yolov3 in fruit detection has been widely studied. Aiming at the problem that Yolov3 loses information during network transmission and the semantic feature extraction of small targets is not rich, this article proposed an improved Yolov3 cherry tomato detection algorithm. Firstly, the proposed algorithm uses dual path network as a feature extraction network to extract richer small target semantic features. Second, four feature layers with different scales are established for multiscale prediction. Finally, the improved K-means++ clustering algorithm is used to calculate the scale of anchor boxes.Experiments showed that the algorithm has a precision rate of 94.29%, a recall rate of 94.07%, and an F1 value of 94.18%. The F1 value is 1.54% higher than Faster R-CNN and 3.45% higher than Yolov3. It takes 58 ms on average to recognize an image, which provides a theoretical basis for the fruit detection.

show abstract

“…The first layer consists of 512 hidden units and the activation function is the ReLU, while the second layer consists of 6 units, as the number of the target classes, with the activation function to be the softmax [19]. Average Pooling [19] and Flattening methods [19] have been used for the frames as to be in a suitable format to be given as input to the layers. Dropout is applied between the two hidden layers.…”

Section: Low Motion Magnitude Activities Modelmentioning

confidence: 99%

“…The new model receives a sequence of frames and it outputs a vector with the probabilities of the classes. The optimization algorithm that has been used is the Stochastic Gradient Descent (SGD) [19] with learning rate 0.0001.…”

Section: Low Motion Magnitude Activities Modelmentioning

confidence: 99%

Real Time Human Activity Recognition Using Acceleration and First-Person Camera data

Androutsos

Tachos

Tripoliti

et al. 2021

2021 43rd Annual International Conference of the IEEE Engineering in Medicine &Amp; Biology Society (EMBC)

View full text Add to dashboard Cite

The aim of this work is to present an automated method, working in real time, for human activity recognition based on acceleration and first-person camera data. A Long-Short-Term-Memory (LSTM) model has been built for recognizing locomotive activities (i.e. walking, sitting, standing, going upstairs, going downstairs) from acceleration data, while a ResNet model is employed for the recognition of stationary activities (i.e. eating, reading, writing, watching TV working on PC). The outcome of the two models is fused in order for the final decision, regarding the performed activity, to be made. For the training, testing and evaluation of the proposed models, a publicly available dataset and an "in-house" dataset are utilized. The overall accuracy of the proposed algorithmic pipeline reaches 87.8%.

show abstract

A deep neural network approach towards real-time on-branch fruit recognition for precision horticulture

Cited by 58 publications

References 34 publications

Multi-Target Recognition of Bananas and Automatic Positioning for the Inflorescence Axis Cutting Point

Multi-Target Recognition of Bananas and Automatic Positioning for the Inflorescence Axis Cutting Point

An improved Yolov3 based on dual path network for cherry tomatoes detection

Real Time Human Activity Recognition Using Acceleration and First-Person Camera data

Contact Info

Product

Resources

About