Improved Kiwifruit Detection Using Pre-Trained VGG16 With RGB and NIR Information Fusion
Bibliographic record
Abstract
This study presents a novel method to apply the RGB-D (Red Green Blue-Depth) sensors and fuse aligned RGB and NIR images with deep convolutional neural networks (CNN) for fruit detection. It aims to build a more accurate, faster, and more reliable fruit detection system, which is a vital element for fruit yield estimation and automated harvesting. Recent work in deep neural networks has led to the development of a state-of-the-art object detector termed Faster Region-based CNN (Faster R-CNN). A common Faster R-CNN network VGG16 was adopted through transfer learning, for the task of kiwifruit detection using imagery obtained from two modalities: RGB (red, green, blue) and Near-Infrared (NIR) images. Kinect v2 was used to take a bottom view of the kiwifruit canopy's NIR and RGB images. The NIR (1 channel) and RGB images (3 channels) were aligned and arranged side by side into a 6-channel image. The input layer of the VGG16 was modified to receive the 6-channel image. Two different fusion methods were used to extract features: Image-Fusion (fusion of the RGB and NIR images on input layer) and Feature-Fusion (fusion of feature maps of two VGG16 networks where the RGB and NIR images were input respectively). The improved networks were trained end-to-end using back-propagation and stochastic gradient descent techniques and compared to original VGG16 networks with RGB and NIR image input only. Results showed that the average precision (APs) of the original VGG16 with RGB and NIR image input only were 88.4% and 89.2% respectively, the 6-channel VGG16 using the Feature-Fusion method reached 90.5%, while that using the Image-Fusion method reached the highest AP of 90.7% and the fastest detection speed of 0.134 s/image. The results indicated that the proposed kiwifruit detection approach shows a potential for better fruit detection.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".