Category: 3. Business

Get ready to hear a lot about Jerome Powell.

Remarks from the Federal Reserve chair will likely demand market watchers’ attention this week as investors seek clarity over the central bank’s next interest-rate moves. Powell has been under pressure to produce rate cuts, but recent economic data has put officials in a tough position.

Traders also will be following earnings expected from major retailers, including Walmart, Target, Home Depot, Lowe’s, and Ross Stores. Investors will watch for signs of tariff-driven inflation and fading consumer sentiment. Housing market data, Fed meeting minutes, and weekly jobless claims also could have an impact on markets this week.

The major U.S. indexes logged gains last week, with the Dow touching an intraday record on Friday.

Read to the bottom for our calendar of key events—and one more thing.

Investors Look to Powell Remarks for Clarity on Interest- Rate Path

Attention will turn toward the American West this week. At the annual Jackson Hole Economic Policy Symposium, Powell is expected to lead a lineup of speakers that includes central bankers, economists, and top officials. 

Economists are seeing more likelihood that the Fed will cut interest rates at its next meeting as the central bank faces relentless pressure from President Donald Trump and other administration officials to lower borrowing costs when it next meets in September. The Fed hasn’t lowered rates since last December and now finds itself in a tough position, said BMO Senior Economist Jennifer Lee, with inflation ticking higher while the job market looks weaker than thought. 

“Can’t imagine the pressure on Fed Chair Powell ahead of the Jackson Hole gathering,” Lee wrote in a recent blog post.

The minutes for the July meeting of the Federal Open Market Committee will provide a look into the Fed’s view of interest rates and the economy and could add insight about the actions of two committee members who split from their colleagues to vote in favor of a rate cut last month.

Housing market data and jobless claims also will be released this week.

Walmart, Target Earnings Due as Tariff Pressures Loom

As Trump’s tariffs begin to show some impact on inflation, earnings reports from large retailers will show if the import taxes are hitting their sales. 

Walmart’s (WMT) scheduled report on Thursday comes after the retailer said it would look to price increases to help balance the costs of tariffs. Home Depot (HD) has said it was attempting to maintain its pricing structure, with investors getting more details on the hardware retailer with its report set for Tuesday. 

Target’s (TGT) expected Wednesday report follows warnings in the prior quarter that sales may move lower than originally projected. Sales also seen softening for T.J. Maxx parent TJX (TJX), which said in the prior quarter that tariffs are expected to impact revenue figures. Other noteworthy retailers scheuled to report this week include home improvement chain Lowe’s (LOW) and discount retailer Ross Stores (ROST).

Quick Links: Recap Last Week’s Trading | Read Investopedia’s Latest News

This Week’s Calendar

Monday, Aug. 18

• Homebuilder confidence (August)
• Key Earnings: Palo Alto Networks (PANW)

Tuesday, Aug. 19

• Housing starts (July)
• Key Earnings: Home Depot, Medtronic (MDT), Keysight Technologies (KEYS), Viking Holdings (VIK), XPeng (XPEV), Toll Brothers (TOL)

Wednesday, Aug. 20

• FOMC minutes for July meeting
• Fed Officials Speaking: Atlanta Fed President Raphael Bostic
• Key Earnings: TJX Cos., Lowe’s, Analog Devices (ADI), Target, Estee Lauder (EL), Baidu (BIDU)

Thursday, Aug. 21

• Existing home sales (July)
• Fed Officials Speaking: Atlanta Fed President Raphael Bostic, Jackson Hole Economic Policy Symposium begins
• Key Earnings: Walmart, Intuit (INTU), Workday (WDAY), Ross Stores
• Data to Watch: Initial jobless claims (Week ending Aug. 16), Philadelphia Fed manufacturing survey (August), S&P Flash U.S. PMI (August), U.S. leading economic indicators (July)

Friday, Aug. 22

• Expected remarks from Fed Chair Jerome Powell at Jackson Hole symposium
• Key Earnings: BJ’s Wholesale Club (BJ)

One More Thing

Social Security celebrates its 90th anniversary this month, but fewer Americans are confident in the program’s long-term financial stability. Investopedia’s Aaron McDade has more on the potential benefit reductions that the program faces.

This summer, Russia’s hackers put a new twist on the barrage of phishing emails sent to Ukrainians.

The hackers included an attachment containing an artificial intelligence program. If installed, it would automatically search the victims’ computers for sensitive files to send back to Moscow.

That campaign, detailed in July in technical reports from the Ukrainian government and several cybersecurity companies, is the first known instance of Russian intelligence being caught building malicious code with large language models (LLMs), the type of AI chatbots that have become ubiquitous in corporate culture.

Those Russian spies are not alone. In recent months, hackers of seemingly every stripe — cybercriminals, spies, researchers and corporate defenders alike — have started including AI tools into their work.

LLMs, like ChatGPT, are still error-prone. But they have become remarkably adept at processing language instructions and at translating plain language into computer code, or identifying and summarizing documents.

The technology has so far not revolutionized hacking by turning complete novices into experts, nor has it allowed would-be cyberterrorists to shut down the electric grid. But it’s making skilled hackers better and faster. Cybersecurity firms and researchers are using AI now, too — feeding into an escalating cat-and-mouse game between offensive hackers who find and exploit software flaws and the defenders who try to fix them first.

“It’s the beginning of the beginning. Maybe moving towards the middle of the beginning,” said Heather Adkins, Google’s vice president of security engineering.

In 2024, Adkins’ team started on a project to use Google’s LLM, Gemini, to hunt for important software vulnerabilities, or bugs, before criminal hackers could find them. Earlier this month, Adkins announced that her team had so far discovered at least 20 important overlooked bugs in commonly used software and alerted companies so they can fix them. That process is ongoing.

None of the vulnerabilities have been shocking or something only a machine could have discovered, she said. But the process is simply faster with an AI. “I haven’t seen anybody find something novel,” she said. “It’s just kind of doing what we already know how to do. But that will advance.”

Adam Meyers, a senior vice president at the cybersecurity company CrowdStrike, said that not only is his company using AI to help people who think they’ve been hacked, he sees increasing evidence of its use from the Chinese, Russian, Iranian and criminal hackers that his company tracks.

“The more advanced adversaries are using it to their advantage,” he said. “We’re seeing more and more of it every single day,” he told NBC News.

The shift is only starting to catch up with hype that has permeated the cybersecurity and AI industries for years, especially since ChatGPT was introduced to the public in 2022. Those tools haven’t always proved effective, and some cybersecurity researchers have complained about would-be hackers falling for fake vulnerability findings generated with AI.

Scammers and social engineers — the people in hacking operations who pretend to be someone else, or who write convincing phishing emails — have been using LLMs to seem more convincing since at least 2024.

But using AI to directly hack targets is only just starting to actually take off, said Will Pearce, the CEO of DreadNode, one of a handful of new security companies that specialize in hacking using LLMs.

The reason, he said, is simple: The technology has finally started to catch up to expectations.

“The technology and the models are all really good at this point,” he said.

Less than two years ago, automated AI hacking tools would need significant tinkering to do their job properly, but they are now far more adept, Pearce told NBC News.

Another startup built to hack using AI, Xbow, made history in June by becoming the first AI to climb to the top of the HackerOne U.S. leaderboard, a live scoreboard of hackers around the world that since 2016 has kept tabs on the hackers identifying the most important vulnerabilities and giving them bragging rights. Last week, HackerOne added a new category for groups automating AI hacking tools to distinguish them from individual human researchers. Xbow still leads that.

Hackers and cybersecurity professionals have not settled whether AI will ultimately help attackers or defenders more. But at the moment, defense appears to be winning.

Alexei Bulazel, the senior cyber director at the White House National Security Council, said at a panel at the Def Con hacker conference in Las Vegas last week that the trend will hold, at least as long as the U.S. holds most of the world’s most advanced tech companies.

“I very strongly believe that AI will be more advantageous for defenders than offense,” Bulazel said.

He noted that hackers finding extremely disruptive flaws in a major U.S. tech company is rare, and that criminals often break into computers by finding small, overlooked flaws in smaller companies that don’t have elite cybersecurity teams. AI is particularly helpful in discovering those bugs before criminals do, he said.

“The types of things that AI is better at — identifying vulnerabilities in a low cost, easy way — really democratizes access to vulnerability information,” Bulazel said.

That trend may not hold as the technology evolves, however. One reason is that there is so far no free-to-use automatic hacking tool, or penetration tester, that incorporates AI. Such tools are already widely available online, nominally as programs that test for flaws in practices used by criminal hackers.

If one incorporates an advanced LLM and it becomes freely available, it likely will mean open season on smaller companies’ programs, Google’s Adkins said.

“I think it’s also reasonable to assume that at some point someone will release [such a tool],” she said. “That’s the point at which I think it becomes a little dangerous.”

Meyers, of CrowdStrike, said that the rise of agentic AI — tools that conduct more complex tasks, like both writing and sending emails or executing code that programs — could prove a major cybersecurity risk.

“Agentic AI is really AI that can take action on your behalf, right? That will become the next insider threat, because, as organizations have these agentic AI deployed, they don’t have built-in guardrails to stop somebody from abusing it,” he said.

Bitcoin’s price consolidation that started a few days ago continues, as the asset seems stuck at the $118,000 level.

While most larger-cap alts have mimicked BTC’s underwhelming performance during the weekend, some, such as OKB, MNT, XMR, and LINK, have soared by double digits.

BTC Consolidation Endures

The business week started on the right foot for bitcoin as the bulls initiated a leg up that pushed it from $118,000 to just over $122,000 on Monday alone. Although the asset retraced in the following days, it went on the offensive hard on Wednesday and Thursday morning once again.

The culmination occurred in the early hours of August 14, when BTC skyrocketed past its July all-time high and set a new one at just over $124,500. Following this $5,000-$6,000 price pump in hours, though, came the almost inevitable correction that drove BTC down to $121,000.

The worst was yet to come later that day, after the release of the hot PPI data for July. Bitcoin reacted with an immediate price drop to under $118,000, dragging the altcoins with it. Since then, BTC has remained sideways at around $118,000 even though there was a big but disappointing meeting between the presidents of the US and Russia.

For now, its market cap stands still at $2.350 trillion, while its dominance over the alts is down to 57.6% on CG.

Double-Digit Pumps for These Alts

As the graph below will show, most larger-cap alts are with minor gains or losses over the past day. ETH, BNB, SOL, DOGE, HYPE, XLM, and SUI are slightly in the green, while XRP, ADA, and TRX have charted insignificant losses.

In contrast, LINK has soared by over 10% in the past day and now sits above $24. XMR and MNT have also charted similar gains, while OKB has stolen the show once again with a 17% surge that has driven it to over $120.

The cumulative market cap of all crypto assets has added over $30 billion since yesterday and is up to $4.080 trillion on CG.

Taipei, Aug. 17 (CNA) Though the government recently raised its 2025 GDP growth forecast to 4.45 percent due to stronger-than-expected AI-driven exports, economists warned that American tariffs and weak domestic consumption could hurt growth momentum in the coming months.

The United States imposed a baseline tariff of 20 percent on goods made in Taiwan that took effect on Aug. 7, higher than the 15 percent tariffs imposed on goods made in Japan and South Korea, raising concern over the tariff’s impact on Taiwan’s economy.

Gordon Sun (孫明德), director of the Taiwan Institute of Economic Research (TIER) Economic Forecasting Center, said U.S. tariff policy remained a major uncertainty for Taiwan’s economy, and the government needed to clearly assess its economic impact.

Sun felt, however, that Taiwan’s exports would remain stable in the coming months because 70 percent of its exports are ICT products, which he said were currently unaffected by the U.S. tariffs.

The other 30 percent are exports of products from non-tech industries, which have come under pressure, Sun said, but most of those exports tend to go to China and Southeast Asia, not the U.S., and therefore should not be too badly affected.

In addition, the government has also introduced support programs for exporters and subsidies for affected industries, which he described as “insurance” to soften the blow.

The bigger economic challenge for Taiwan in the coming months is domestic demand, Sun argued.

Retail sales fell 0.4 percent year-on-year in the first half, and 2.9 percent in June alone, he said, citing data from the Ministry of Economic Affairs.

He cited two reasons for weak consumption — uncertainty over the tariffs, which has delayed car purchases, and a cooling property market after government measures against speculation, which has hurt consumer confidence.

To boost confidence, the government must take more proactive measures, such as universal cash handouts to stimulate demand and monetary easing by the central bank, Sun said.

Dachrahn Wu (吳大任), director of the National Central University (NCU) Research Center for Taiwan Economic Development, warned that U.S. President Donald Trump’s tariff policy is aimed at forcing companies to invest and produce in the U.S.

That could squeeze domestic investment and drive high-paying jobs overseas, severely damaging Taiwan’s domestic demand, Wu said.

Reshaping supply chains

Chiou Jiunn-rong (邱俊榮), an economics professor at NCU, argued that businesses must be mentally prepared for a long struggle as tariffs have become the new normal, regardless of U.S. leadership.

He said Trump’s broader tariff war could restructure supply chains and alter global business cycles, slowing growth for years, and while subsidies could help affected industries in the short term, they risked leaving uncompetitive industries stagnant.

In the longer run, Taiwan needs to consolidate its strength in semiconductors and also reinforce sectors less exposed to tariffs, including applications, software and smart systems, to maintain competitiveness amid economic uncertainty, Chiou said.

The scholars’ comments came after the Directorate-General of Budget, Accounting and Statistics (DGBAS) Department of Statistics on Friday sharply raised its 2025 GDP growth forecast from 3.1 percent in May to 4.45 percent.

It attributed the upward revision to a surge in exports and private investment driven by strong AI demand.

The DGBAS, however, revised the growth of private consumption down to 0.85 percent. Of the 4.45 percent projected gains in GDP, net external demand contributed 2.71 percentage points of that, while private consumption contributed only 0.4 percentage points, the DGBAS said.

The experimental process

To enhance the model’s generalization ability and convergence speed, pretrained backbone network parameters were used. During training, the input image size was set to 480 × 480, and the number of epochs was adjusted to 100. The optimal combination of hyperparameters and the best optimization algorithm were selected to ensure the model could effectively learn a sufficient number of features. The Adam optimizer was used for all models, with an initial learning rate of 0.001. If the loss showed little variation, the learning rate was reduced by 50%.

Table 5 indicates that the selection of hyperparameters significantly affects the segmentation performance of the HRNet model in apple leaf disease segmentation tasks.

The performance of the HRNet model demonstrates remarkable stability across different hyperparameter settings, with minimal variations in evaluation metrics. The results indicate that HRNet is relatively stable when applied to the apple leaf disease segmentation task. The optimal segmentation performance is achieved when the learning rate is set to 0.0001 and the Adam batch size is 8.

To evaluate the segmentation performance of the Improved HRNet model for four types of apple leaf diseases, a comparison with the original HRNet model was made. Figure 7 compares the performance of HRNet and Improved HRNet in terms of IoU and PA for four diseases: Alternaria Blotch, Brown Spot, Grey Spot, and Rust.

The results show that, compared to HRNet, the Improved HRNet significantly improved in the Alternaria Blotch segmentation task, with an 11.54% point increase in IoU and a 7.98% point increase in PA. In the Brown Spot segmentation task, IoU increased by 1.5% points and PA by 0.38% points. In the Grey Spot segmentation task, IoU increased by 9.14% points and PA by 2.6% points. However, in the Rust segmentation task, IoU increased by 2.71% points and PA by 0.65% points. These results indicate that the introduction of the NAM attention mechanism in the Improved HRNet model enhances the focus on local features of the lesions, improving the effectiveness of feature extraction and, consequently, the segmentation accuracy. On the other hand, the Brown Spot and Rust diseases have more distinctive features, so HRNet already performed well in these categories, resulting in a smaller improvement in the Improved HRNet model for these tasks.

Comparison of different backbone networks

This study improved the HRNet model by testing three different backbone network widths: HRNet_w18, HRNet_w32, and HRNet_w48, to compare their performance in apple leaf disease segmentation. Figures 8(a) and 8(b) present the training results for apple leaf disease image segmentation using models with different backbone network widths. The experimental results are shown in Table 6.

The results show that HRNet_w32 provides the best overall performance. Its average IoU (mIoU) is 82.21%, which is an improvement of 2.07% points over HRNet_w18 and 0.07% points over HRNet_w48. The average pixel accuracy (mPA) is 89.59%, which is an improvement of 2.71% points over HRNet_w18 and 0.97% points over HRNet_w48. Additionally, its average precision (mPrecision) reaches 89.53%. While HRNet_w48 achieves the highest precision of 91.92%, the improvements in mIoU and mPA are marginal. This is likely due to the excessively large network width of HRNet_w48, which leads to overfitting during training and causes a performance bottleneck on the test set. Furthermore, the larger network increases computational complexity and cost, resulting in slightly underwhelming performance in terms of mIoU and mPA. Overall, HRNet_w32 strikes an optimal balance between performance and computational complexity, avoiding the overfitting issue observed with HRNet_w48 while delivering a solid segmentation performance. Therefore, HRNet_w32 is selected as the backbone network for apple leaf disease segmentation tasks.

Comparison of different attention mechanisms

To analyze the performance of different attention mechanisms in the apple leaf disease segmentation task, we conducted comparative experiments using CBAM, SENet, and NAM attention mechanisms. In each experiment, the attention module was added at the same position in the encoder. Figures 9(a) and 9(b) display the training outcomes for models with various attention mechanisms applied to apple leaf disease image segmentation. The experimental results are presented in Table 7.

As shown in the table, when the SENet module was added, the mIoU increased by 0.4%, reaching 85.22%. With the addition of the CBAM module, mIoU improved by 2.54%, reaching 87.36%. The introduction of the NAM module resulted in the highest increase of 3.51%, with an mIoU of 88.33%. This demonstrates that NAM outperforms the other attention modules in terms of segmentation accuracy. NAM enhances the model’s ability to learn apple leaf disease features by effectively weighting multi-source information and reducing the interference of background noise and redundancy. Furthermore, during the feature fusion process across different resolutions, NAM improves the quality of multi-scale feature integration, thereby enhancing overall segmentation performance. Compared to CBAM and SENet, NAM exhibits superior capability in capturing local details, suppressing background noise, and adapting to multi-scale features, which significantly boosts the model’s segmentation performance. In apple leaf disease segmentation tasks, the NAM attention mechanism proves to be the most suitable choice.

Performance analysis of ablation experiment

To systematically evaluate the impact of each module on the overall model performance, we designed four ablation experiments. These experiments assess the effects of replacing the backbone network, adding the attention mechanism, and using Focal Loss and Dice Loss functions. The experimental results are summarized in Table 8.

The table shows that using HRNet_w32 as the backbone network significantly improved segmentation performance, with mIoU and mPA increasing by 4.68 and 5.22% points, respectively. The introduction of the NAM attention mechanism further boosted mIoU and mPA by 3.51 and 1% point, respectively, due to NAM’s enhancement in the multi-scale feature fusion process, which better refines features across different resolutions. The use of Focal Loss effectively addressed the issue of class imbalance, improving mIoU and mPA by 0.47 and 0.17% points, respectively. Dice Loss enhanced segmentation accuracy for small targets and imbalanced classes, with mIoU and mPA increasing by 0.55 and 0.21% points, respectively.

When HRNet_w32, NAM attention mechanism, Focal Loss, and Dice Loss were combined, the model achieved the highest performance, with mIoU and mPA improving by 8.77 and 7.25% points, respectively. This significantly enhanced the model’s segmentation performance for apple leaf disease.

Comparison with other segmentation methods

To further validate the segmentation performance of the improved HRNet model, we compared it with several classic semantic segmentation models commonly used for plant disease tasks, including DeeplabV3 +[²⁶] , U-Net²⁷and PSPNet²⁸. The results are presented in Fig. 10.

As shown in Fig. 9, the proposed model outperforms the others in disease segmentation, achieving the best accuracy with an mIoU of 88.91% and an mPA of 94.13%. The DeeplabV3 + model performed the worst, with an mIoU of 79.20% and an mPA of 87.35%. The U-Net model showed relatively superior segmentation performance, with an mIoU of 80.85% and an mPA of 86.38%. The PSPNet model had an mIoU of 79.71% and an mPA of 87.86%. The experimental results indicate that the NAM attention mechanism incorporated into HRNet enhances the model’s feature extraction and representation abilities. Additionally, the optimization of the loss functions im-proves the model’s segmentation accuracy for diseased areas and addresses the seg-mentation accuracy issues caused by data sample imbalance during training. Overall, the HRNet model, with its high-resolution feature representation, is better suited to the requirements of apple leaf disease segmentation tasks.

This study visualized the segmentation results of five algorithms: Improved HRNet, HRNet, DeeplabV3+, U-Net, and PSPNet, as shown in Fig. 11.

Figure 11 reveals distinct performance variations among models in disease segmentation tasks. The morphological and chromatic similarity between Alternaria Blotch and Grey spot lesions induced misclassification errors in Models C, D, and E, which erroneously identified Alternaria Blotch as Grey spot. These models also demonstrated inadequate precision in segmenting overlapping healthy leaf regions. In Brown Spot segmentation, Models D and E showed minor false positives, while Models B, C, and E suffered significant under-segmentation issues. Grey spot detection revealed two critical failures: Models D and E produced misclassifications, Models B, D, and E generated oversimplified healthy tissue delineation, and Model C even segmented non-existent targets. For Rust identification, Model E exhibited false positives, while Models B-D displayed insufficient resolution in overlapping healthy leaf areas.

Notably, the Improved HRNet achieved accurate four-disease differentiation with exceptional edge delineation and complete lesion morphology while achieving pixel-level precision at disease-leaf boundaries. This architecture demonstrated superior robustness and segmentation accuracy through its hierarchical feature integration mechanism, effectively addressing the critical challenges of inter-class similarity and complex edge topology that compromised conventional models.

Assessment of disease severity levels

To accurately assess the severity of apple leaf diseases, this study refers to the local standard of Shanxi Province, “DB14/T 143–2019 Apple Brown Spot Disease Monitoring and Survey Guidelines,” to establish grading parameters for apple brown spot disease. Based on pixel statistics, Python was used to calculate the pixel count of the diseased and healthy leaf areas. The leaf disease severity was classified into six levels: Level 0 (healthy leaf), Level 1, Level 3, Level 5, Level 7, and Level 9. The detailed leaf disease grading standards are shown in Table 9.

Where, represents the ratio of the diseased area to the area of a single leaf, and is calculated using the following formula:

In the formula, A_scab denotes the area of the diseased region, A_leaf represents the area of a single leaf, and pixel(x, y) is used to count the number of pixels corresponding to the diseased and leaf regions, respectively.

In the process of grading apple leaf diseases in complex backgrounds, the diversity of leaf shapes and the complexity of the background affect pixel statistics, which in turn influences the grading results. To address this, pixel statistical analyses were performed under three scenarios: a single leaf, separated multiple leaves, and overlapping multiple leaves. The DRL-Watershed algorithm was used to accurately count the pixels of the disease and the leaf area in each scenario, ensuring the accuracy of the grading results. The visualized segmentation results of the DRL-Watershed algorithm for the three cases are shown in Fig. 12:

Pixel statistical analysis for a single leaf

To verify the effectiveness of the DRL-Watershed algorithm in pixel counting for a single leaf, a comparative experiment was conducted using the pixel statistics from the improved HRNet model. The grading results for disease severity on a single leaf are shown in Table 10.

As shown in Table 10, for the improved HRNet model, the total number of pixels in the leaf area (sum of leaf and disease pixels) is 127,917, with the disease occupying 46% of the area, resulting in a disease level of Level 9. In the DRL-Watershed algorithm, the number of leaf pixels is 126,026, with the disease area occupying 47%, and the disease level is also Level 9. This demonstrates that both the Improved HRNet model and the DRL-Watershed algorithm were able to accurately count the leaf pixels and calculate the disease proportion, yielding corresponding disease severity levels.

Pixel statistical analysis for separated multiple leaves

The principle of using the watershed algorithm for handling multi-leaf separation in disease severity assessment is illustrated in Fig. 13. To assess the performance of the DRL-Watershed algorithm for scenarios with multiple separated leaves, pixel statistics were compared with the results from the improved HRNet model. The disease severity grading results for the separated leaves are shown in Table 11.

As shown in Table 11, the improved HRNet model calculates the disease-to-leaf pixel ratio across the entire image, resulting in a disease proportion of 18.22% and a disease level of Level 5, which reduces the overall disease proportion and severity. In contrast, the DRL-Watershed algorithm separately counts the pixels for each individual leaf and computes the disease ratio for each leaf, providing a more accurate reflection of the disease severity. For example, the DRL-Watershed algorithm calculates that the disease proportion for Leaf Area 2 is 26.26%, corresponding to a disease level of Level 7, which accurately represents the disease severity on each leaf and offers a more precise grading assessment.

Pixel statistical analysis for overlapping multiple leaves

Similarly, the principle of applying the watershed algorithm for assessing disease severity in overlapping multi-leaf scenarios is demonstrated in Fig. 14. To evaluate the DRL-Watershed algorithm’s performance for scenarios with overlapping leaves, a comparative experiment was performed using pixel statistics from the improved HRNet model. The disease grading results for the overlapping multiple leaves are shown in Table 12.

As shown in Table 12, the improved HRNet model calculates the disease-to-total-leaf pixel ratio, yielding a disease proportion of 24% and a corresponding disease level of Level 5. This method results in an underestimation of the disease severity because the disease pixels are compared to the total leaf area across all leaves. However, the DRL-Watershed algorithm effectively segments the overlapping leaf regions, allowing it to calculate the pixel count for each individual leaf. In the two overlapping areas, the DRL-Watershed algorithm calculates the disease proportion for Area 1 as 32%, with a corresponding disease level of Level 7, and for Area 2 as 22%, corresponding to Level 5. This approach provides a more accurate reflection of the disease severity in each overlapping leaf region, yielding a grading assessment closer to the actual situation.

Severity grading statistical analysis

In this study, the performance of the HRNet model and the DRL-Watershed algorithm in grading apple leaf diseases on the test set was evaluated and analyzed using confusion matrices. The true severity levels were determined based on the ratio of the disease area to the leaf area during the data annotation process. Figure 15 compares the predicted results of the improved HRNet model with the true results, and the disease severity evaluation results of the DRL-Watershed algorithm on the same test set, respectively. The vertical axis represents the true labels, while the horizontal axis represents the model’s predictions. Each cell in the matrix contains the number of samples where a true category was predicted as a specific category. Higher values in the diagonal blocks indicate that the model correctly predicted the disease severity levels. The diagonal elements of the confusion matrix represent the number of correctly classified samples, while the off-diagonal elements represent misclassifications. The intensity of the diagonal color corresponds to the accuracy of the grading for each level.

In the confusion matrix of the HRNet model, 89 samples were correctly classified as Level 1. For Level 3, 21 samples were misclassified as Level 1, while 36 samples were correctly classified as Level 3. In Level 5, 2 samples were misclassified as Level 3, but most samples were correctly classified as Level 5. Levels 7 and 9 exhibited some classification confusion, particularly with samples from Level 7 being misclassified as either Level 5 or Level 9. The HRNet model demonstrates high accuracy in predicting lower disease severity levels. However, there is some error in predicting higher severity levels, which can be attributed to the HRNet model’s tendency to underestimate disease severity in multi-leaf scenarios. This occurs because the HRNet model does not distinguish between individual leaves when processing multiple leaves, leading to lower predicted severity levels compared to the actual severity.

The confusion matrix for the DRL-Watershed algorithm shows significant improvements, particularly for Level 3. In the 55 samples for Level 3, only 2 were misclassified as Level 1, with the rest correctly classified as Level 3. Classification accuracy for Level 5 also improved, with 31 samples correctly classified and only 1 misclassified as Level 3. For Level 9, all 13 samples were correctly classified. Compared to Figure a, the classification results for Level 7 were notably better, with 8 samples correctly classified and no significant misclassifications. These results suggest that the DRL-Watershed algorithm, by separately analyzing the disease proportion in each leaf region, provides a more accurate assessment of disease severity, especially in complex and overlapping leaf scenarios.

In the confusion matrix for the DRL-Watershed algorithm, two Level 3 samples were misclassified as Level 1, one Level 5 sample as Level 3, and one Level 9 sample as Level 3. These errors may be caused by noise introduced by lighting, shadows, or other environmental factors, which result in unclear boundaries in the overlapping leaf regions. In these regions, the gradients are less pronounced, leading to inaccurate seg-mentation of the leaf area, which in turn affects the final disease severity predictions.