Lightweight tea bud recognition network integrating GhostNet and YOLOv5

Abstract

<abstract> <p>Aiming at the problems of low detection accuracy and slow speed caused by the complex background of tea sprouts and the small target size, this paper proposes a tea bud detection algorithm integrating GhostNet and YOLOv5. To reduce parameters, the GhostNet module is specially introduced to shorten the detection speed. A coordinated attention mechanism is then added to the backbone layer to enhance the feature extraction ability of the model. A bi-directional feature pyramid network (BiFPN) is used in the neck layer of feature fusion to increase the fusion between shallow and deep networks to improve the detection accuracy of small objects. Efficient intersection over union (EIOU) is used as a localization loss to improve the detection accuracy in the end. The experimental results show that the precision of GhostNet-YOLOv5 is 76.31%, which is 1.31, 4.83, and 3.59% higher than that of Faster RCNN, YOLOv5 and YOLOv5-Lite respectively. By comparing the actual detection effects of GhostNet-YOLOv5 and YOLOv5 algorithm on buds in different quantities, different shooting angles, and different illumination angles, and taking F1 score as the evaluation value, the results show that GhostNet-YOLOv5 is 7.84, 2.88, and 3.81% higher than YOLOv5 algorithm in these three different environments.</p> </abstract>

Health to Eat: A Smart Plate with Food Recognition, Classification, and Weight Measurement for Type-2 Diabetic Mellitus Patients’ Nutrition Control

“…CNN is made up of an input layer, an output layer, and several hidden layers. The YOLO architecture is made up of only two layers: a convolution layer and a pooling layer [ 40 ].…”

TransLearn-YOLOx: Improved-YOLO with Transfer Learning for Fast and Accurate Multiclass UAV Detection

“…In a dataset of 4792 images [9], 4.6k images are for model training (95%) and 240 (5%) for model validation. All experiments are separately run on the Google Colab environment with an NVIDIA Tesla T4 GPU having a low learning rate of 0.001, where GAANet and GhostNet-YOLOv5 [24] has a batch size of 256 and 512, respectively. The epochs for both models are set to 500 for both models where GhostNet-YOLOv5 [24] stopped training at 300 by using early stopping as the model performance stopped improving while GAANet stopped training at 457 epochs.…”

“…The detailed evaluation of both trained models GAANet and GhostNet-YOLOv5 [24] is performed by the comparison of true positive (TP), true negative (TN), false negative (FN), false positive (FP), mAP, precision, and recall values. The GAANet model achieved the highest TP value of 1.00 for drones and planes and lowest TP of 0.72 for helicopters.…”

“…In [25], the authors improved the baseline YOLOv4 backbone with GhostNet and achieved 0.9% less precision than GAANet. Similarly, the authors used Ghost convolution in the YOLOv5 baseline model and achieved 19.89% less precision than GAANet [24]. YOLOv5x-ALL- GHOST added GhostNet in both head and backbone and got 0.6% less map@0.5 than GAANet [26].…”

High-Accuracy Insulator Defect Detection for Overhead Transmission Lines Based on Improved YOLOv5

“…To solve the problems of insufficient accuracy and lack of robustness in the process of insulator defect fault detection, this paper proposes a YOLOv5 based on a receptive field module and multiscale. The main work is as follows: anchor frames are obtained that match the size of the detected target by k-means clustering to improve the detection accuracy of the network for target objects with different proportions; the low-level detail features are extracted from the network and fused with the deepest semantic features to the small-scale detection layer designed in this paper to improve the detection performance of the network model for small-area targets; a lightweight backbone network is built using the GhostNet [26] lightweight network to reduce convolution operations and improve the real-time performance of the model while ensuring detection accuracy; the channel receptive field block (CRF) receptive field module that integrates channel information is designed at the network head to replace the original SPP module [27], integrate channel information, fuse multiscale feature information, and use dilated convolution to reduce the calculation of redundant information.…”