23 March 2020 in IEEE Access
To improve the recognition model accuracy of crop disease leaves and locating diseased leaves, this paper proposes an improved Faster RCNN to detect healthy tomato leaves and four diseases: powdery mildew, blight, leaf mold fungus and ToMV. First, we use a depth residual network to replace VGG16 for image feature extraction so we can obtain deeper disease features. Second, the k-means clustering algorithm is used to cluster the bounding boxes. We improve the anchoring according to the clustering results. The improved anchor frame tends toward the real bounding box of the dataset. Finally, we carry out a k-means experiment with three kinds of different feature extraction networks. The experimental results show that the improved method for crop leaf disease detection had 2.71% higher recognition accuracy and a faster detection speed than the original Faster RCNN.
30 July 2019 in IEEE Access
StampedLock, proposed in JDK 1.8, provides many interesting features, such as optimistic read locks and upgrading/downgrading locks to improve the design of concurrent programs instead of employing pure read/write locks. Existing refactorings have proposed algorithms to convert locks, but there are a few refactorings that use these promising features of StampedLock. To illustrate a possible refactoring, this paper first shows three code transformations based on StampedLock. Then, this paper presents CLOCK, an automated refactoring tool that helps developers convert the synchronized lock into the StampedLock. An algorithm for reentrance analysis is proposed for the precondition validation. The write lock, read lock, optimistic read lock, and upgrading/downgrading lock are inferred and refactored. CLOCK is evaluated with the SPECjbb2005 benchmark and two real-world applications, Xalan and FOP. A total of 66 classes are modified by searching approximately 395KSLOC and applying the refactoring, achieving an average of 22 classes per benchmark. The experimental results show that CLOCK can help developers with refactoring for StampedLock and save developer effort.
1 March 2018 in Neurocomputing
28 September 2021 in IEEE Access
Go provides a new concurrency mechanism based on message-passing, which brings a series of new concurrency bugs. The existing tools are highly dependent on model-checking methods in which they take the source program as the input of model-checking tools to detect the potential concurrent bugs. However, these methods can only qualitatively obtain the concurrency of the program but not obtain the exact number of concurrency bugs in the program. Meanwhile, these approaches are not sensitive to dead codes, which are prone to false negatives. Furthermore, few works have been done on detecting WaitGroup-related bugs in Go concurrent programming, which may cause deadlock. To this end, this paper proposes a novel approach GoDetector to detect concurrent bugs. GoDetector uses a detecting algorithm and a pushdown automaton to verify the channel-related concurrent bugs and WaitGroup-related concurrent bugs separately. To support the evaluation of the tool, 30 benchmarks are collected from the existing tools. GoDetector reports 2 channel safety errors, 29 deadlocks, and 11 WaitGroup-related errors. Compared with the existing tool Godel, GoDetector has the capacity to report more concurrency bugs and fewer false positives. GoDetector also offers additional support for the detection of WaitGroup variables. The experimental results indicate that GoDetector is more efficient at detecting the concurrency in regards to accuracy and diversity.