Core i7 12700

This experiment deploys python 3.19.13 and torch-1.11.0+cu113 deep learning framework for model training and testing on a computer with an Intel Core i7-12700 CPU @ 64-bit 4.90 GHz, 32 GB RAM, NVIDIA GeForce RTX 3060 12045MiB GPU graphics card and ubuntu22.04LTS operating system. Table 1 provides the details of all the trained models. Using the general procedures for network training on YOLOv5 platform, the proposed lightweight YOLOv5-LiNet including other YOLO related models takes an input of 512×512 pixels, 16 batch size, 0.937 momentum, 0.0005 weight decay, 0.2 IoU, 0.015 hue, 0.7 saturation, 0.4 lightness, 1.0 mosaic and 300 epochs training, while Mask-RCNN received an input of 512×512 pixels with default parameters on MMdetection platform. Random initialization technique was used to initialize the weights for training all the models from scratch.

Pearson correlation coefficient was used to determine the relationship between two gradient results and all gradient results. The SciPy package in Python was used to calculate the Pearson correlation coefficient, where the parameters $p$ and $r$ reflect the significance level and correlation, respectively. Significant differences were considered when $p<0.05$ . The positive or negative value of $r$ reflected the positive or negative correlation. Difference analysis was used to analyze the differences of different gradient results, as well as to analyze the effect of data augmentation methods.
All the experiments were run on Intel Core i7-12700 (2.10 GHz), 64GB RAM and NVIDIA RTX3060 GPU with 12 GB memory. Python 3.7.1 was used to implement all program code. The functions in the Sckit-learn package were used to realize the KNN and SVM algorithms. The MobileViT-xs model was implemented by the deep learning framework PyTorch 1.10, and CUDA Toolkit 11.3 was used to accelerate the processing. More code details were posted at https://github.com/mepleleo/DA_peanut (accessed on 1 April 2022).