Geforce gtx 1070ti gpu

For data analysis, two workstation computers were used. Both systems boot into Windows 10 (Microsoft) off a 1 TB M.2 drive (Samsung 970 EVO Plus). The first system consists of an I9-9900X processor (Intel), a GeForce GTX 1070 Ti GPU (Nvidia) and 128 GB of DDR4 RAM (Corsair). The second system has an I9-9900K processor (Intel), a GeForce RTX 2070 GPU (Nvidia) and 64 GB of DDR4 RAM (G.Skill Ripjaws). Data were stored on a 4 TB RAID0 array consisting of two 2 TB drives (Samsung) and a 2 TB RAID0 array consisting of two 1 TB Drives (Samsung), respectively. System integration, support and maintenance performed by Nobska Imaging, Inc.

The experiments of this research were conducted using an Intel i5 64-bit quad-core CPUs operating at a frequency of 3.30 GHz (Santa Clara, CA, USA). The system had 16 GB of RAM and an NVIDIA GeForce GTX 1070Ti GPU with 8 GB memory. The chosen model framework was PyTorch, with CUDA 11.1 and Python 3.8.10 for implementation. Table 1 lists some hyper parameters used in the experiments.
The criteria used for assessing the performance of fruit detection encompassed precision, recall, mean average precision (mAP), and F₁ score (Padilla et al., 2020 (link)). The metrics are defined in (Equations 9–12):
where R and P are the recall and precision, respectively. Using mAP is a valuable approach to assess the model performance across different confidence levels.
with AP expresses in (Equation 11.a):
where $p(\tilde{r})$ represents the calculated precision at a given recall value ( $\tilde{r}$ ), while $N_{cls}$ is the total number of classes.

In this study, the experiments were conducted on a computer that has Intel i5 (Santa Clara, CA, USA), 64-bit 3.30 GHz quad-core CPUs, and a NVIDIA GeForce GTX 1070Ti GPU.
The model receives images of $416\times 416$ pixels as inputs. Due to GPU memory constraints, the batch size was set to 8. The model was trained for 160 epochs with an initial learning rate of ${10}^{-3}$ , which was then divided by 10 after 60 and 90 epochs. The momentum and weight decay were set to 0.9 and 0.0005, respectively.
A series of experiments were conducted to evaluate the performance of the proposed method. The indexes for evaluation of the trained model are defined as follows: $Recall=\frac{TP}{TP+FN}$
$Precision=\frac{TP}{TP+FP}$
where TP, FN, and FP are abbreviations for true positives (correct detection), false negatives (miss), and false positives (false detection).
To better show the comprehensive performance of the model, F₁ score was adopted as a trade-off between the recall and precision, defined in Equation (15): $F_1=\frac{2\times Recall\times Precision}{Recall+Precision}$
Another evaluation metric for object detection—Average Precision (AP) [34 (link),36 (link)]—was also used in this study. It can show the overall performance of a model under different confidence thresholds, and is defined as follows: $AP=\sum _n\left(r_{n+1}-r_n\right)p_{interp}\left(r_{n+1}\right)$
with
$p_{interp}\left(r_{n+1}\right)=\underset{\tilde{r}:\tilde{r}\ge r_{n+1}}{max}p\left(\tilde{r}\right)$
where $p\left(\tilde{r}\right)$ is the measured precision at recall $\tilde{r}$ .