Physiological　data　are　nowadays　frequently　used　to　recognize　the　affective　state　of　subjects　while　performing　different　tasks.　Measuring　cognitive　load　using　simple　device　plays　an　important　role　in　everyday　life　such　as　intelligent　human-computer　interaction,　physical　health　monitoring,　and　mental　health　monitoring.　However,　due　to　the　nature　of　the　experiments　involving　subjects,　the　obtained　data　base　is　often　low,　making　it　difficult　to　train　deep　learning　methods　from　scratch.　In　this　paper,　We　conducted　a　method　to　recognize　the　cognitive　load　based　on　the　Photoplethysmogram(PPG)　data　and　the　application　of　Long　Short　Term　Memory　(LSTM).　We　tested　this　method　on　the　PPG　data　from　an　experiment　where　19　subjects　were　involved　in　arithmetic　calculation　tasks　of　two　different　cognitive　load　levels.　The　method　successfully　achieve　btter　accuracy　of　recognition　compared　with　the　traditional　machine　learning　classifiers　with　features　artificially　extracted　from　the　image　and　pre-trained　CNN　method,　92.3%　binary　classification　accuracy　was　reached　and　about　3.8%　binary　classification　accuracy　was　improved.　　©　2023　ACM.