GOAL: For the current traditional high-resolution to low-resolution to high-resolution feature extraction form, there will be some problems, such as the U-net network model skip connections, this method of reducing the low-level features of the encoder directly to the high-level features of the decoder will destroy the original high-level semantic features and lead to excessive segmentation.
MAIN RESEARCH CONTENT AND METHODS; EXPERIMENTAL PLAN
约束：1000字以内，宋体五号，单倍行距，1页以内，打印The dataset for this experiment will be using brain tumor images extracted from Kaggle Website, for preprocessing. The preprocessing step raises the standard of brain tumor MRl images and prepares them for upcoming processing by health experts, helps with signal-to-noise ratio, removal of noise and background of unwanted parts, smoothing regions, and maintaining relevant edges. segmentation will be used in dividing an image into different regions. Region growing is used in the process of combining pixels or subregions into larger regions based on specific criteria. The image is segmented, and the segmented image is used to identify the desired tumor region, which then will be later be detected.
- Divide the available data into training and test subsets.
- Iterate through the optimization loop a set number of times or until a condition is met
- Evaluate all metric values and select the hyperparameter set that produces the best metric value
To achieve this problem a topic of Medical image detection of Efficientnets in parallel structure was proposed.